Advertisement

Climate Dynamics

, Volume 49, Issue 9–10, pp 3011–3029 | Cite as

Timeslice experiments for understanding regional climate projections: applications to the tropical hydrological cycle and European winter circulation

  • Robin Chadwick
  • Hervé Douville
  • Christopher B. Skinner
Article

Abstract

A set of atmosphere-only timeslice experiments are described, designed to examine the processes that cause regional climate change and inter-model uncertainty in coupled climate model responses to \(CO_2\) forcing. The timeslice experiments are able to reproduce the pattern of regional climate change in the coupled models, and are applied here to two cases where inter-model uncertainty in future projections is large: the tropical hydrological cycle, and European winter circulation. In tropical forest regions, the plant physiological effect is the largest cause of hydrological cycle change in the two models that represent this process. This suggests that the CMIP5 ensemble mean may be underestimating the magnitude of water cycle change in these regions, due to the inclusion of models without the plant effect. SST pattern change is the dominant cause of precipitation and circulation change over the tropical oceans, and also appears to contribute to inter-model uncertainty in precipitation change over tropical land regions. Over Europe and the North Atlantic, uniform SST increases drive a poleward shift of the storm-track. However this does not consistently translate into an overall polewards storm-track shift, due to large circulation responses to SST pattern change, which varies across the models. Coupled model SST biases influence regional rainfall projections in regions such as the Maritime Continent, and so projections in these regions should be treated with caution.

Keywords

Regional climate change Climate model uncertainty Tropical hydrological cycle European winter circulation 

1 Introduction

Climate model projections of future climate change are very uncertain in many regions, particularly for the atmospheric circulation and related quantities such as precipitation (Collins et al. 2013; Shepherd 2014; Kent et al. 2015; Xie et al. 2015; Oueslati et al. 2016; Long et al. 2016). Uncertainty across models can be due to differences in forcing (scenario uncertainty), differences in their responses to forcing (model uncertainty), or internal variability (Hawkins and Sutton 2011). Of these three factors, model uncertainty should be able to be reduced by either improving all climate models, or by determining which subset of climate models is likely to give the most realistic response to forcing for a given region. Both of these strategies rely on advancing our understanding of the processes that drive regional climate change, and how these processes differ across models.

In this study we present a new set of experiments, designed to examine the mechanisms that cause regional climate change in coupled General Circulation Models (GCMs), and inter-model uncertainty in this change. These experiments will form part of Coupled Model Intercomparison Project phase 6 (CMIP6). As examples, we apply this methodology to two regions where future climate change is uncertain across models, and the processes that drive change are not well understood: (1) the tropical hydrological cycle, and (2) winter circulation over the North Atlantic and Europe.

A common approach to achieving adequate signal-to-noise when examining mechanisms of regional climate change is to focus on high emissions scenarios such as Representative Concentration Pathway 8.5 (RCP8.5). RCP scenarios include future changes in greenhouse gases, atmospheric aerosols and land-use, all of which can produce regional climate responses. However at very high greenhouse gas forcing levels, such as are present in RCP8.5, the pattern of regional climate change is very similar to that in experiments where only \(CO_2\) is increased (abrupt4x\(CO2\) experiments), suggesting that the RCP8.5 response is dominated by greenhouse gases (Kent et al. 2015). For simplicity in this study we consider only regional climate responses to \(CO_2,\) which are likely to dominate high emissions scenarios, though the responses to aerosol and land-use forcing will also need to be considered for lower emissions scenarios.

The processes that combine to produce regional climate change in a coupled model can be separated using atmosphere-only GCM (AGCM) experiments. These can be run with atmospheric constituents changed (e.g. \(CO_2\) increased) but sea-surface temperatures (SSTs) fixed, with an imposed uniform or patterned SST change, or with some combination of the two. This approach has been widely used in individual GCMs to examine the relative contributions of direct forcing change, SST warming, and sometimes SST pattern change across a large variety of different regions and variables (Folland 1998; Cash et al. 2005; Douville 2005; Deser and Phillips 2009; Dong et al. 2009; Gastineau et al. 2009; Xie et al. 2010; Skinner et al. 2012; Ma et al. 2012; Harvey et al. 2015; Shaw and Voigt 2015; Dong and Sutton 2015; Chen and Zhou 2015; Zhang and Li 2016). More recently, this has been extended to a multi-model context using idealised experiments with the CMIP5 (Taylor et al. 2012) ensemble (Biasutti 2013; Ma and Xie 2013; Bony et al. 2013; Chadwick et al. 2014; Thorpe and Andrews 2014; He et al. 2014; Shaw and Voigt 2015; Voigt and Shaw 2015; He and Soden 2015a; Richardson et al. 2016; Chadwick 2016; Gaetani et al. 2016; Kamae et al. 2016).

Analyses of CMIP5 idealised AGCM amip (present-day SSTs, sea-ice and atmospheric constituents), amip4K (uniform 4K SST warming), amip4xCO2 (SSTs fixed, \(CO_2\) quadrupled), and amipFuture (4K pattern SST warming) experiments have produced useful insights into the mechanisms at work in coupled models, which will be discussed later in more detail. However these amip experiments were not designed to reproduce coupled model projections at regional scales, and so suffer from a number of deficiencies when used to understand regional climate change in coupled GCMs (Chadwick 2016). The sum of amipFuture and amip4xCO2 precipitation change in each model is not spatially well correlated with abrupt4x\(CO2\) or RCP8.5 projections, so it is not possible to quantitatively decompose the coupled precipitation response into responses to SST warming, SST pattern change and increased \(CO_2\) in most regions (Chadwick 2016). This is demonstrated for three CMIP5 models in Fig. 1.
Fig. 1

Climatological annual mean precipitation change (\({\rm mm}\, {\rm day}^{-1}\)) in the abrupt4xCO2, a4SST-4xCO2, and amipFuture + amip4xCO2 experiments for each of the three models. Spatial correlation coefficients between the coupled abrupt4xCO2 precipitation change and each of the other experiments are also shown

The lack of ocean–atmosphere coupling in AGCM experiments is well known to cause significant problems in their representation of internal variability (e.g. Barsugli and Battisti 1998; Wu and Kirtmann 2004) and can also lead to errors in their projections of mean future climate change (Douville 2005). However, the differences between future change in the CMIP5 coupled and amip experiments (Fig. 1) are not mainly due to the lack of coupling, but are instead due to the particular experimental design of the amip experiments. The base-state SSTs are different between coupled and amip simulations, as are the prescribed patterns of future SST change—in the case of amipFuture the same SST pattern anomaly is applied to all models, whereas each coupled model produces its own pattern of SST change. Additionally, at least one important process that is present in coupled models is not present in the idealised amip simulations—the plant physiological response to \(CO_2\) forcing (e.g. Betts et al. 2008), which is not active in the amip4xCO2 experiments.

When AGCM experiments are forced with SSTs and atmospheric constituents directly from present-day and future sections (known as timeslices) of a coupled run with increasing \(CO_2\) forcing, they are capable of closely reproducing the mean coupled model change between these timeslices (Skinner et al. 2012; He and Soden 2015b)—though some discrepancies do remain (Douville 2005). Similarly, when forced with observed SSTs and atmospheric constituents, AGCMs can reproduce observed regional trends in circulation, precipitation and temperature (Deser and Phillips 2009; Shin and Sardeshmukh 2011). Therefore a framework of AGCM timeslice experiments, forced with the various individual and combined aspects of SST warming and \(CO_2\) forcing from coupled climate simulations, should be capable of isolating the various mechanisms of climate change that take place in coupled models on a region by region basis. Such a framework will form part of the Cloud Feedbacks Model Intercomparison Project (CFMIP) contribution to CMIP6 (Webb et al. 2016). An additional experiment is also included, in order to examine the influence of present-day SST biases on future climate projections. This experiment uses the same future SST and \(CO_2\) anomalies as in the timeslice experiments, but they are imposed on top of an amip (present-day observed) SST base-state.

A pilot study of these experiments has been performed with three CMIP5 models. The methodology and several examples of regional applications are described here. The aim is to decompose regional climate responses in a coupled abrupt4x\(CO2\) experiment into the parts associated with uniform SST warming, SST pattern change, the direct effect of increased \(CO_2,\) and the plant physiological response to increased \(CO_2.\) In the final CMIP6 experimental design, the influence of sea-ice change will also be isolated. This timeslice framework allows a number of research questions to be asked:
  1. 1.

    How do regional climate responses (e.g. of precipitation) in a coupled model arise from the combination of responses to different aspects of \(CO_2\) forcing (direct \(CO_2\) effect, plant physiological effect) and sea surface warming (uniform SST warming, patterned SST warming)?

     
  2. 2.

    Which aspects of forcing/warming are most important for causing inter-model uncertainty in regional climate projections?

     
  3. 3.

    Can inter-model differences in regional projections be related to underlying structural or resolution differences between models through improved process understanding, and could this help us to constrain the range of regional projections?

     
  4. 4.

    What impact do coupled model SST biases have on regional climate projections?

     
Although all these experiments are primarily designed for examining time-mean climate responses, they are also likely to be useful for understanding changes on other timescales, such as daily extremes, though in these cases the lack of air-sea coupling could be influential.

We perform a detailed analysis of two regions where future model uncertainty is large—the tropics and the North Atlantic/European region—in order to demonstrate the utility of the timeslice experiments. We focus on hydrological cycle and atmospheric circulation change in these regions, as these are both particularly uncertain and crucially important for assessing potential climate change impacts. As well as providing a proof-of-concept for the forthcoming CMIP6 experiments, the 3-model results shown here provide some interesting insights into the dominant mechanisms of climate change in each region. In particular, the high importance of the plant physiological effect in tropical forest regions is demonstrated.

Section 2 describes the experimental design used in the pilot study, and a comparison between regional responses to \(CO_2\) in coupled, timeslice and amip experiments in the three models. This is followed by a description of minor modifications that will be made for the final CMIP6 experimental design. Section 3 examines hydrological cycle change in the tropics, and Sect. 4 describes circulation and precipitation change over the North Atlantic and Europe. Section 5 examines the influence of present-day model SST biases on future regional climate change. A summary and conclusions are presented in Sect. 6.

2 Experimental design

The CMIP5 experiments used here are (1) piControl: a coupled ocean–atmosphere control run with pre-industrial forcing. (2) abrupt4xCO2: as piControl, but with \(CO_2\) concentrations abruptly quadrupled at the start of the experiment and then held constant. (3) amip (Atmosphere Model Intercomparison Project): an atmosphere-only experiment, with observed forcings and SST boundary conditions from 1979–2008. (4) amip4xCO2: as amip, but with \(CO_2\) quadrupled throughout the experiment, though SST boundary conditions are the same as for amip. (5) amipFuture: as amip, but with an imposed climatological SST pattern anomaly taken from the ensemble mean of CMIP3 1%/year \(CO_2\) increase runs at the time of \(CO_2\) quadrupling, scaled to have a global mean of +4K. The applied SST pattern anomaly is the same for all models.
Table 1

SST boundary conditions and \(CO_2\) concentrations used in each of the timeslice experiments

Experiment

SSTs

\(CO_2\)

piSST

Years 101–130 of piControl

amip 1979–2008

piSST-p4K

As piSST, with uniform +4K anomaly

amip

piSST-4xCO2-rad

As piSST

4x amip, with increase seen only by radiation

piSST-4xCO2

As piSST

4x amip

a4SST

As piSST, with patterned anomaly from abrupt4xCO2-piControl

amip

a4SST-4xCO2

As a4SST

4x amip

amip-a4SST-4xCO2

1979-2008 amip SSTs, with the same anomaly as a4SST

4x amip

The new timeslice experiments (Table 1) are (1) piSST: the same as amip but with monthly-varying SSTs and sea-ice from years 101–130 of each models own control run rather than observed fields. (2) piSST-p4K: the same as piSST but with SSTs uniformly increased by 4K. Sea-ice is unchanged from piSST. (3) piSST-4xCO2-rad: the same as piSST but \(CO_2\) as seen by the radiation scheme (but not the plant physiological effect) is quadrupled. (4) piSST-4xCO2: same as piSST but with \(CO_2\) quadrupled, and this increase is seen by both the radiation scheme and the plant physiological effect. (5) a4SST: the same as piSST, but a patterned SST anomaly is applied on top of the monthly-varying piSST SSTs. This anomaly is a monthly climatology of the difference between each model’s own abrupt4xCO2 and piControl runs (using the mean of years 91–140 of abrupt\(CO2,\) and the parallel 50 years section of piControl), then scaled to have a global mean of 4K. Sea-ice is unchanged from piSST. (6) a4SST-4xCO2: the same as a4SST, but \(CO_2\) is also quadrupled, and is seen by both the radiation scheme and the plant physiological effect. Sea-ice is unchanged from piSST. (7) amip-a4SST-4xCO2: as amip, but with the SST pattern anomaly climatology from a4SST added to the amip observed SSTs. CO\(_2\) is quadrupled, and is seen by both the radiation scheme and the plant physiological effect. Sea-ice is unchanged from amip.

a4SST-4xCO2 is used to establish whether a timeslice experiment can adequately recreate the coupled abrupt4xCO2 response in each model, and then forms the basis for a decomposition using the other experiments. The influence of SST pattern change was examined by subtracting piSST-p4K from a4SST. The response to the plant physiological effect was isolated by subtracting piSST-4xCO2-rad from piSST-4xCO2. Finally the non-linear interaction between SST change and CO\(_2\) change was estimated by taking the difference between (a4SST-4xCO2 − piSST) and (piSST-4xCO2 + a4SST − 2 x piSST). Table 2 shows how each component of change is calculated.
Table 2

Definition of components of \(CO_2\) forcing and warming from the piSST experiments

Component

Definition

All forcings

a4SST-4xCO2 − piSST

SST pattern-only

a4SST − piSST-p4K

SST uniform +4K

piSST-p4K − piSST

4x\(CO_2\) Rad.

piSST-4xCO2-rad − piSST

Veg. response to 4x\(CO_2\)

piSST-4xCO2 − piSST-4xCO2-rad

Nonlinear response

a4SST-4xCO2 − a4SST − piSST-4xCO2 + piSST

The three coupled models on which this study was based are CCSM4 (Meehl et al. 2012), CNRM-CM5 (Voldoire et al. 2011) and HadGEM2-ES (Martin et al. 2011). For simplicity, we refer hereafter to the ACGM configurations of these coupled models as CCSM4, CNRM-CM5 and HadGEM2. In the case of the CNRM-CM5 runs, there was a minor difference from the experimental design described above. Sea-ice was not fixed at piSST or amip values in the piSST-p4K, a4SST, a4SST-4xCO2 and amip-a4SST-4xCO2 experiments, but was allowed to respond to the warming SSTs. SST anomalies (either +4K or abrupt4xCO2–piControl) were applied at all ocean and sea-ice points in CNRM-CM5, which then re-classified points as sea-ice if the temperature was below −1.8 \(^{\circ }\)C, or as open ocean above this threshold. The model version of CNRM-CM5 used to run the timeslice experiments was also very slightly different to that used for the CMIP5 model experiments (including piControl and abrupt4xCO2), with a minor change to the calculation of the bulk coefficients for surface turbulent fluxes. This does not make a large difference to the control or future climate simulated by the model, but is one possible reason why the agreement between coupled and timeslice anomalies is not quite so close for CNRM-CM5 as for the other two models.

All models were regridded to a 2.5\(^{\circ }\) grid for comparison with one another. In order to examine the response to forcing, 30-year mean differences between experiments were used. For amip4xCO2, amipFuture and the new timeslice experiments, the difference with amip or piSST over the period 1979–2008 was used. For abrupt4xCO2, the difference with piControl over years 91–140 was used. The abrupt4xCO2–piControl anomaly was scaled by a factor of 4K/(global mean SST warming) for each model, so that the global mean change was 4K in each case.

Multi-decadal internal variability of precipitation was estimated in each of the three models, in order to determine how influential this might be in the timeslice results. 240 years of each model’s piControl experiment was split into consecutive 30-year periods, then the standard deviation (\(\sigma \)) of precipitation across these periods was calculated at each gridpoint. A value of \({\sqrt 2}\sigma\) was used as the expected internal variability for anomalies between future and control experiments.

2.1 Final CMIP6 experimental design

After analysis of these pilot experiments, some small changes were made for the final CMIP6 experimental design of the timeslice experiments. The final protocol is described in Webb et al. (2016), and that document should be referred to by modelling groups interested in running these experiments for CMIP6. The main changes are the addition of an extra experiment to examine the influence of sea-ice changes, the use of piControl atmospheric constituents (instead of amip values) in piSST, and the use of actual abrupt4xCO2 monthly-and-annually-varying SSTs in a4SST instead of applying a monthly climatology anomaly to the piSST SSTs—making the final design more of a pure timeslice approach. These modifications should improve even further the agreement between the coupled and AGCM results.

2.2 Global comparison between coupled and AGCM responses to forcing

Figure 1 shows the precipitation response to forcing in the abrupt4xCO2, a4SST-4xCO2 and amipFuture + amip4xCO2 experiments with each of the models. Agreement with the coupled responses is much better in the new timeslice experiments (spatial correlation coefficients are shown in Fig. 1) than in the amip experiments, and the timeslice responses capture both the pattern and magnitude of coupled responses across the majority of regions. This suggests that the new timeslice experimental framework can be used to decompose the coupled model responses into their different aspects of forcing and warming at regional scales.

3 Regional changes in the tropical circulation and hydrological cycle

CMIP5 models consistently project that large, societally important changes in tropical rainfall will occur in response to CO\(_2\) forcing (Chadwick et al. 2015), but disagree on the location of these changes (McSweeney and Jones 2013; Collins et al. 2013; Kent et al. 2015; Oueslati et al. 2016; Long et al. 2016). Here, we examine each of the mechanisms that contributes to regional tropical rainfall (Figs. 2, 3, 4), circulation (Fig. 5) and soil moisture (Fig. 6) change, and how the balance between mechanisms varies by region. The timeslice experiments reproduce the total pattern of coupled model change well in this region (Figs. 2a, b, 3a, b, 4a, b) and so can be used to reliably decompose the change into different mechanisms, though the agreement is not perfect (Figs. 2c, 3c, 4c).
Fig. 2

HadGEM2-ES tropical annual mean precipitation change in a abrupt4xCO2 and bh components of change calculated from the timeslice experiments (\({\rm mm} \, {\rm day}^{-1}\)). i An estimate of 30 years multidecadal internal precipitation variability in the anomalies \(({\rm mm} \, {\rm day}^{-1}).\) j SST pattern change (anomaly from tropical mean change) in abrupt4xCO2 (K)

Fig. 3

As Fig. 2 but for CCSM4

Fig. 4

As Fig. 2 but for CNRM-CM5. Note that the plant physiological response is not included in CNRM-CM5

Fig. 5

HadGEM2-ES tropical annual mean circulation change for a abrupt4xCO2 and bh components of change calculated from the timeslice experiments. Colours show change in 500 hPa vertical pressure velocity (\({\rm Pa} \, {\rm s}^{-1}\)), and arrows show change in horizontal 925 hPa winds (\({\rm m} \, {\rm s}^{-1}\))

Fig. 6

Tropical annual mean total soil moisture change (\({\rm kg} \, {\rm m}^{-2}\)) in a, c, e abrupt4xCO2 and b, d from the plant physiological response for the two models that include this process

3.1 The plant physiological effect, and its importance in tropical forest regions

Plant stomata narrow in response to increased \(CO_2\) concentrations (known as the plant physiological response), leading to reduced transpiration and feedbacks on circulation, precipitation and other parts of the hydrological cycle (e.g. Betts et al. 2008). Recently, Boisier (2017) used a set of coupled model experiments, with and without the vegetation response to \(CO_2\) turned on, to demonstrate that the plant physiological response is the dominant driver of water cycle change over the Amazon region in CMIP5 projections. The timeslice experiments used here confirm this result, and also show how the plant physiological response combines with other drivers of precipitation change in tropical forest regions. The new AGCM experiments have the additional advantage of being much computationally cheaper to run than coupled model experiments.

It should be noted that the new AGCM experiments do not recreate any changes in vegetation location/type that may occur in coupled GCMs. As the AGCM runs are still able to successfully recreate the same pattern of change as seen in the full GCMs, this suggests that vegetation type changes are not major influences on the variables examined in this study, though they may have some stronger regional effects. Leaf area index (LAI) was able to respond in the HadGEM2 AGCM experiments, but not in the other two models. However, the results of Skinner et al. (2016) suggest that the LAI effect on hydrological cycle and circulation change is small compared to that of the plant stomatal response.

HadGEM2-ES and CCSM4 both represent the plant physiological effect (Figs. 2g, 3g) but CNRM-CM5 does not (Fig. 4g—it will be included in the CMIP6 CNRM model). Over the Amazon the plant physiological response is the single biggest factor in precipitation change for both CCSM4 and HadGEM2-ES, whereas the total Amazon rainfall change in CNRM-CM5 is much weaker in the absence of the plant response. This is consistent with Boisier (2017), who found that models without the plant physiological response have much weaker Amazon water cycle responses, and suggested that the CMIP5 ensemble mean response may therefore underestimate the true extent of water cycle change. The pattern of rainfall change in response to the plant physiological effect is complex (Figs. 2g, 3g), and is not simply a reduction in rainfall over all tropical forest regions in response to reduced transpiration. This suggests that dynamical feedbacks on transpiration change play a significant role in determining the pattern of rainfall change, and this can be seen in Fig. 5g.

Over the Amazon, CCSM4 and HadGEM2-ES both show a north-east south-west dipole of rainfall decreases and increases in response to the plant physiological effect. Over the Maritime Continent, both models show strong rainfall increases, possibly driven by land warming caused by reductions in transpiration, and a resulting increased land–sea temperature gradient. Again, this is crucial to the overall rainfall increases in this region in these two models, and CNRM-CM5 does not show large rainfall increases over the Maritime Continent. CCSM4 also has a strong response over Africa, with a west-east dipole of precipitation changes, but this is not present in HadGEM2-ES.

The plant physiological effect is also important for other aspects of water cycle change, such as runoff (Betts et al. 2007) and soil moisture. Figure 6 shows the total change in soil moisture in each of the three models, and the contribution from the plant physiological effect in each case. The plant response leads to soil moisture increases across most of tropical land, associated with reduced moisture loss via transpiration. Exceptions are regions where the plant response leads to large rainfall reductions in either model. The inclusion of the plant response is likely to be important for determining the total change in soil moisture, as changes are much smaller in CNRM-CM5 than in the other two models.

3.2 SST pattern change

SST pattern change has previously been identified as a crucial source of inter-model uncertainty over the tropical oceans (Xie et al. 2010; Chadwick et al. 2013; Ma and Xie 2013; Kent et al. 2015; Oueslati et al. 2016; Chadwick 2016; Long et al. 2016), either by taking the residual between coupled model uncertainty and uncertainty in AGCM uniform SST experiments, or by using singular value decomposition methods to relate variations in SST change and precipitation change to one another across models. The influence of SST pattern change on changes over land is much less clear, though a connection has been suggested in several regions (Li et al. 2006; Harris et al. 2008; Giannini et al. 2013). Anderson et al. (2015) suggested that 25% of the inter-model variance in CMIP5 precipitation change over land can be explained by SST pattern change uncertainty. However, analysis of the CMIP5 idealised amip experiments suggested that SST pattern change may not be an important driver of inter-model uncertainty over much of tropical land (Chadwick 2016). As SST pattern change does not in general occur independently of the atmosphere (Xie et al. 2010), the climate response to an imposed SST pattern change may be better interpreted as the response associated with ocean–atmosphere coupling (Long et al. 2016), or at least the part of that response that can be captured within AGCM experiments. The new timeslice experiments allow the influence of each individual model’s SST pattern change (shown in Figs. 2j, 3j, 4j) to be examined in more detail.

Over the tropical oceans, the response to SST pattern change is the most important mechanism of overall precipitation pattern change in HadGEM2 (spatial correlation coefficient between the SST pattern component and total change is 0.68) and CCSM4 (0.71). This is reflected in the pattern of circulation change (Fig. 5a, d), with low-level wind anomalies generally flowing from cooler SSTs towards warmer ones. The relationship between the SST pattern component and total rainfall change is weaker in the west Pacific and South Pacific Convergence Zone (SPCZ) regions, where the influence of uniform SST increases is very large.

In CNRM-CM5 the gradients of SST pattern change are not as strong, and consequently the SST pattern response is not as highly correlated with the total response (0.34). For this model, the overall oceanic response is a combination of responses to SST patterns, uniform warming and direct CO\(_2\) increases. The weaker SST pattern change in CNRM-CM5, combined with the lack of plant physiological response in this model (see Sect. 3.1), explains why its overall precipitation pattern response is muted compared to the other two models. One simple measure of this is a comparison of the spatial standard deviations of the SST pattern change and precipitation change (over tropical ocean) fields in each model. These are (0.68, 0.49 and 0.29 K) and (1.32, 1.25 and 0.62 \({\rm mm} \, {\rm day}^{-1}\)) for HadGEM2, CCSM4, and CNRM-CM5 respectively. In the CMIP5 (amipFuture + amip4xCO2) experiments (Fig. 1), where all three of the models are forced with the same pattern of SST pattern change, the magnitude of CNRM-CM5 precipitation change is not noticeably weaker than that of the other two models. Overall, the results of this three-model ensemble suggest that SST pattern change is a very important contributor to both mean rainfall change and inter-model uncertainty over the tropical oceans, but that it is not dominant in every model.

Over land, the influence of SST pattern change varies by region and by model. In general, the magnitude of terrestrial precipitation change driven by SST pattern change is consistent with the strength of SST pattern change in each model, as it is over the oceans. HadGEM2 and CCSM4 exhibit large responses over tropical land, particularly over South America, where rainfall anomalies are likely to be associated with Atlantic ocean ITCZ shifts in each model (see also Joetzjer et al. 2013). All three models simulate a substantial precipitation response to SST pattern change in Central America. In other regions, the SST pattern response is mixed across the three models. For example, a strong east-west dipole of precipitation change over Africa is present in CCSM4, but not in the other two models.

In general, these results suggest that SST pattern change is likely to contribute to inter-model uncertainty in rainfall change over land, in agreement with Anderson et al. (2015), and that the results of Chadwick (2016) may underestimate the influence of SST pattern change due to limitations of the CMIP5 amip experiments. In this case, model uncertainty in hydrological cycle changes over land may be able to be reduced by constraining future SST pattern change.

3.3 Responses to uniform warming and direct \(CO_2\) forcing

Uniform SST warming tends to reduce rainfall over land and enhance it over the oceans, whereas the direct \(CO_2\) effect tends to enhance rainfall over land and reduce it over the oceans (Figs. 2e, f, 3e, f, 4e, f). The two effects partially cancel over much of the tropics, and in some regions such as West Africa the total rainfall change is strongly influenced by the shape of the residual between the two terms (Giannini 2010; Skinner et al. 2012; Biasutti 2013; Gaetani et al. 2016). Over land, this has been described as the competing effects of local \(CO_2\) forcing and remote SST warming on rainfall change (Giannini 2010).

SST warming results in increased atmospheric moisture, and this leads to a general thermodynamic increase in rainfall across tropical wet regions (e.g. Held and Soden 2006; Seager et al. 2010). However, the largest cause of regional rainfall change in amip4K experiments is spatial shifts in the pattern of convection (Chadwick 2016), particularly a general movement of convection from land to ocean, and this also appears to apply to the piSST-p4K experiments (Fig. 5e, f). Several processes can drive these dynamical rainfall changes in response to uniform SST warming, including warming aloft over land in response to increased oceanic rainfall (Held et al. 2005; Giannini 2010), increased land–sea temperature gradients (Bayr and Dommenget 2013), and associated decreases in relative humidity over land (Fasullo 2012).

The direct \(CO_2\) effect comprises two distinct mechanisms: a direct atmospheric effect, and a land-warming effect. Atmospheric stabilisation due to enhanced long-wave absorption by the extra \(CO_2,\) with a warming peak at around 700 hPa, acts to suppress convection across the tropics (Sugi and Yoshimura 2004; Andrews et al. 2009; Dong et al. 2009; Cao et al. 2012). As this extra LW absorption is partly ‘masked’ in ascent regions by deep convective clouds and high humidity, but less so in descent regions, the tropical circulation is required to transport less energy and weakens (Merlis 2015). Secondly, the land warms in response to increased downwelling long-wave radiation, and the resultant enhanced land–sea temperature gradient can also drive rainfall changes. The direct atmospheric effect accounts for a substantial part of the overall slow-down of the tropical circulation (Bony et al. 2013), but regional changes in the amip4xCO2 experiments appear to be dominated by the land heating effect (Chadwick et al. 2014; Richardson et al. 2016), a result supported by AGCM experiments where \(CO_2\) concentrations are fixed but the land surface is warmed (Ackerley et al. 2016).

The magnitude of the tropical mean rainfall shifts between land and ocean are quite consistent across the three models. For piSST-p4K, the (mean sea \(\varDelta P -\) mean land \(\varDelta P\)) values are 0.42, 0.37 and 0.49 \({\rm mm} \, {\rm day}^{-1}\) for CNRM-CM5, CCSM4 and HadGEM2, respectively. For piSST-4x\(CO2\)-rad, the (mean land \(\Delta P -\)  mean sea \(\Delta P\)) values are 0.70, 0.78 and 0.70 \({\rm mm} \, {\rm day}^{-1}\) respectively. However, the patterns of change show variation across the three models, and some land regions (such as India) have enhanced precipitation in response to SST warming, while others (such as Central America) have suppressed precipitation in response to increased \(CO_2.\) A detailed analysis of each region (which is beyond the scope of this study) could allow the different model responses—and the residual between piSST-p4K and piSST-4x\(CO2\)-rad—to be linked to underlying differences in model biases or formulation, particularly once a larger ensemble of models have performed these experiments in CMIP6.

In (and only in) regions where SST pattern change is found across models to be unimportant for precipitation change in the timeslice experiments, this may indicate that the use of coupled atmosphere–ocean GCMs is not necessary to produce accurate future projections. For such regions, it may be appropriate to treat the sum of amip4K and amip4xCO2 as the most reliable future projection (although the current design of amip4x\(CO2\) would also not capture the plant physiological effect). In this case the amip experiments should capture the responses to uniform SST warming and direct \(CO_2\) heating without the complicating presence of present-day SST biases, though this approach would still leave substantial uncertainty across models (Chadwick 2016).

3.4 Nonlinear responses

Regional precipitation responses to \(CO_2\) can be strongly nonlinear (Chadwick and Good 2013; Good et al. 2016). This nonlinear precipitation response is partly driven by nonlinear interactions between the climate responses to \(CO_2\) forcing and SST warming (Figs. 2h, 3h, 4h). The nonlinear term is calculated as a residual (see Sect. 2), and so could potentially be influenced by multidecadal internal variability in the timeslice experiments. However, 30-year precipitation variability was estimated from the piControl experiments for each model, and is fairly small (Figs. 2i, 3i, 4i). Therefore, this method is likely to be capturing genuine nonlinear behaviour.

All three models have nonlinear responses, though the regions in which these occur vary by model. HadGEM2-ES has a very strong nonlinear response off the west coast of Central America, which is also present in a much weaker form in CNRM-CM5. CCSM4 has strong nonlinearities over the western tropical Atlantic and central Africa. In general the nonlinear responses are not dominant over the linear terms, and so the timeslice decomposition framework can be considered quasi-linear in the tropics.

4 Changes in North Atlantic and European winter circulation and precipitation

Winter circulation change in the European and North Atlantic region is another area of large uncertainty in climate model projections (Collins et al. 2013; Fereday et al. 2017), and this hinders adaptation planning for future extreme events associated with precipitation and wind-storms. Results from idealised aquaplanet simulations suggest that warming should lead to a poleward shift of the extratropical storm-tracks (e.g. Voigt and Shaw 2015), and this is seen in fully coupled GCMs for the southern hemisphere storm-track (e.g. Bengtssonn and Roeckner 2006). However, the response of the North Atlantic mean circulation and storm-track varies across models (Catto et al. 2011; Cattiaux et al. 2013; Harvey et al. 2014), and the ensemble mean response is better described as an eastward extension rather than a poleward shift (Woollings 2012; Zappa et al. 2013). Here, we examine the drivers of mean changes in North Atlantic and European pressure at mean sea level (PMSL), 200 hPa geopotential height and precipitation in the timeslice experiments. The location of the climatological storm-track in each model was estimated, by calculating the latitude of the maximum mean zonal wind speed at 850 hPa for each longitude band in the region 30–90N, 10–60W (see blue lines on Figs. 7a, b, 8a, b, 9a, b).
Fig. 7

HadGEM2-ES North Atlantic/European DJF mean PMSL (colours hPa) and 200 hPa geopotential height (line contours m) change in a abrupt4xCO2 and bh components of change calculated from the timeslice experiments. i SST pattern change (anomaly from 60N–60S mean change) in abrupt4xCO2 (K), with regions covered by sea-ice masked in white. The blue line is an estimate of the mean storm-track latitude in a piControl and b piSST

Fig. 8

As Fig. 7 but for CCSM4

Fig. 9

As Fig. 7 but for CNRM-CM5. Note that the plant physiological response is not included in CNRM-CM5

A number of mechanisms have been proposed to explain projected circulation changes over this region. Changes in meridional temperature gradients at upper or lower levels can alter the available energy from which individual storms grow, and also affect the related mean winds and pressure patterns (Harvey et al. 2014). Upper level changes in temperature gradient can occur even in response to a uniform surface warming, as the tropical upper troposphere warms more than that of the extra-tropics. Low-level temperature gradient changes can be driven by SST pattern change in the Atlantic, often associated with the Atlantic Meridional Overturning Circulation (AMOC) (Woollings 2012) and intense ocean heat uptake in the deep-water formation regions (Manabe et al. 1990; Xie et al. 2010; Long and Xie 2015), or by decreases in sea-ice coverage or Arctic amplification of low-level temperature increases (Harvey et al. 2015). Changes in latent heating (Schneider et al. 2010) or cloud radiative heating (Voigt and Shaw 2015) in a warmer climate could also alter these temperature gradients. In addition to changes in meridional temperature gradients, increasing static stability in the subtropics can change the local baroclinicity of the atmosphere, shifting the stormtracks polewards. Changes in tropical rainfall, and associated teleconnections, have also been suggested as a possible influence on future change in the North Atlantic Oscillation (NAO) (Deser and Phillips 2009; Fereday et al. 2017). Land heating by the direct \(CO_2\) effect plays a role in circulation change during the summer (Shaw and Voigt 2015), and may also be influential in winter. The influence of the plant physiological effect on circulation change in this region has been less studied, but will also be examined here. Stratosphere–troposphere interactions have also been suggested as a source of uncertainty in future projections over Europe and the North Atlantic (e.g. Scaife et al. 2012), but are not explicitly isolated in these experiments.

Several of these mechanisms could partially or fully offset each other (Harvey et al. 2015; Voigt and Shaw 2015; Shaw and Voigt 2015), and this is likely to be a fundamental cause of model uncertainty in the region. This compensation between mechanisms can also lead to strong nonlinear behaviour at different \(CO_2\) levels (Catto et al. 2011). Most of the processes described here are captured in the timeslice experiments, with the exception of sea-ice change which is not included in the CCSM4 or HadGEM2 experiments, but which will be added as an additional experiment in the final CMIP6 experimental design (Webb et al. 2016). The timeslice experiments are able to qualitatively reproduce the coupled model changes in this region (Figs. 7a, b, 8a, b, 9a, b), and are therefore useful for understanding the components of the coupled response. Differences between the coupled and timeslice experiments are in general larger than for the tropics (Figs. 7c, 8c, 9c), particularly in the Arctic for HadGEM2 and CCSM4, and this could be because of the absence of sea-ice changes in those models. There is also a large difference to the south of Iceland in CNRM-CM5, which could be because of the slightly different model version used in the CNRM-CM5 timeslice experiments (see Sect. 2). Internal multi-decadal variability again appears to be small compared to the forced signal of change (Fig. 10i), so this is not the main cause of differences between the coupled and timeslice experiments.
Fig. 10

HadGEM2-ES North Atlantic/European DJF mean precipitation change in a abrupt4xCO2 and bh components of change calculated from the timeslice experiments (\({\rm mm} \, {\rm day}^{-1}\)). i An estimate of 30 years multidecadal internal precipitation variability in the anomalies (\({\rm mm} \, {\rm day}^{-1}\))

4.1 Response to uniform SST warming

In response to uniform SST warming (Figs. 7e, 8e, 9e), all three models simulate high pressure anomalies in the southern half of the domain, and low pressure anomalies in the northern half of the domain, corresponding to a northward shift of the storm-track. Although the models disagree on the exact magnitude and location of the anomalous low and high pressure centres, their responses to uniform warming are nevertheless fairly consistent. 200 hPa geopotential height anomalies show a similar signal, with larger increases in the southern half of the domain in all three models. This suggests that the reason why climate models do not consistently project a northward shift of the storm-tracks in this region is due to opposing contributions from the responses to other aspects of forcing and warming.

Precipitation responses to uniform warming are consistent with the PMSL anomalies (shown for HadGEM2 in Fig. 10e), with increases in the north of the domain and decreases in the south. The polewards stormtrack shift in these experiments is likely to be associated with increased upper-tropospheric equator-to-pole temperature gradients, and increased subtropical static stability (Catto et al. 2011; Harvey et al. 2014). In CNRM-CM5, sea-ice decreases might also play a role, partially opposing the poleward shift by reducing low-level equator-pole temperature differences (Harvey et al. 2014).

4.2 Response to SST pattern change

North Atlantic SST pattern change (relative to 60N–60S mean SST warming) is shown in Figs. 7i, 8i, 9i, and major differences are seen between the three models. This region warms less than the global mean due to strong ocean heat uptake in the deep-water formation regions, and a future weakening of the AMOC in all models. This AMOC weakening is particularly strong in CCSM4, where a patch of North Atlantic SSTs actually cools slightly, even under high \(CO_2\) forcing (Meehl et al. 2012). Although CNRM-CM5 and HadGEM2 exhibit more similar magnitudes of overall warming in the region, they disagree on the pattern of SST change. These differences are reflected in the PMSL response to SST pattern change (Figs. 7d, 8d, 9d), which is strong in all three models but has a different spatial structure in each. CNRM-CM5 and HadGEM2 both have low pressure anomalies in the south of the domain, and high pressure in the north, with the anomalous low pressure centred over Greenland in CNRM-CM5 and Scandinavia in HadGEM2. In CCSM4 the PMSL response is a strong high pressure anomaly centred to the south of Greenland and Iceland, close to the main SST cool patch in that model. 200 hPa geopotential height changes follow the patterns of PMSL change quite closely in HadGEM2-ES and CNRM-CM5, but less so in CCSM4. Precipitation changes are again consistent with the PMSL anomalies, with HadGEM2 having precipitation reductions in northern Europe and increases in southern Europe.

SST pattern change and uniform SST warming are generally the largest drivers of circulation and precipitation change in these three models, with the response to SST pattern change being far more variable than the response to uniform SST increases across the three. The spatial colocation of SST and PMSL anomalies, particularly in CCSM4, suggest that local SST changes may be playing a large role in driving circulation changes. This supports the analysis of Woollings (2012), which argues that uncertainty in future ocean circulation changes are a major cause of uncertainty in North Atlantic stormtrack changes. However, because the timeslice experiments are not able to separate out the roles of local and remote SST forcing, these circulation responses could also be partly or mainly driven by tropical SST pattern changes, and related changes in tropical rainfall and latent heating (Deser and Phillips 2009; Fereday et al. 2017). These local and remote contributions could potentially be separated using an additional timeslice experiment where a patterned SST anomaly is imposed in the tropics, and a uniform SST anomaly in the extratropics (as in Deser and Phillips 2009). In any case, SST pattern change appears to be the main reason why the North Atlantic stormtrack does not simply shift poleward in these models (see also Grise and Polvani 2014). As in the tropics, it is important to interpret the response to SST pattern change as the response associated with ocean–atmosphere coupling, not simply as being driven by ocean circulation changes.

4.3 Responses to \(CO_2\) and nonlinearity

The magnitude and pattern of the responses to direct \(CO_2\) forcing and the plant physiological effect (Figs. 7f, g, 8f, g, 9f, g) differ between the models. CNRM-CM5 has a strong PMSL response to direct \(CO_2\) forcing, which reinforces its response to uniform SST warming. The other two models have weaker responses, with CCSM4’s direct \(CO_2\) response also reinforcing its uniform SST response, whereas HadGEM2’s direct \(CO_2\) response partially offsets its uniform SST response. As also shown for the summer season (Shaw and Voigt 2015), the North Atlantic/European region is unusual in that the circulation responses to direct \(CO_2\) forcing and uniform SST warming do not in general oppose one another. This may be because uniform SST warming and direct \(CO_2\) forcing do not produce opposing equivalent potential temperature gradients between land and ocean in this region (Shaw and Voigt 2015), though it is unclear why this is the case.

The circulation response to the plant effect is stronger in HadGEM2 than in CCSM4, and in both cases this response partially opposes the responses to direct \(CO_2\) forcing. Physically it is unclear why this should be the case. The plant physiological effect is clearly not as important in this region as it is in tropical forest regions, but could still potentially influence circulation change.

Nonlinear interactions between SST change and increased \(CO_2\) (Figs. 7h, 8h, 9h, 10h) are relatively more important in the North Atlantic region than in the tropics. Interestingly, the nonlinear responses are very different across the three models. Although nonlinear interactions are quite large and should not be neglected, they do not invalidate the linear separation of coupled GCM responses into different mechanisms in this region. Again, analysis of multidecadal precipitation variability in the piControl experiments (Fig. 10i) suggests that these nonlinear terms represent forced nonlinear behaviour rather than internal model variability. Deser and Phillips (2009) also found strong nonlinear interactions between SST change and increased \(CO_2\) in this region, though the nonlinearity in the timeslice experiments is not as large as in their experiments.

5 The impact of SST biases on future precipitation changes

In this section we compare the ‘all-forcings’ timeslice experiment, a4SST-4x\(CO2,\) which uses piControl SSTs and sea-ice as a base-line, with a similar experiment, amip-a4SST-4x\(CO2,\) which uses observed amip SSTs and sea-ice as a baseline. As the two experiments are forced with the same future SST and \(CO_2\) anomalies, any differences between them are due to SST biases in the coupled piControl experiment (and potentially also differences between pre-industrial and present-day SST states). As control SST biases are likely to also affect the future pattern, and possibly magnitude, of SST change in coupled runs, these experiments will not capture the full effect of SST biases on future projections. The lack of coupling in the AGCM experiments could also affect these results. Instead, the difference between a4SST-4x\(CO2\) and amip-a4SST-4x\(CO2\) can be interpreted as showing regions where SST biases are likely to affect the accuracy of future projections in each model. These experiments should also be useful for highlighting regions where future projections could be constrained by considering the present-day SST bias of each model, such as has been performed for precipitation projections in the tropical Pacific (Brown et al. 2015; Huang and Ying 2015).
Fig. 11

Climatological annual mean precipitation change (\({\rm mm} \, {\rm day}^{-1}\)) for a4SST-4xCO2-piSST, and amip-a4SST-4xCO2-amip, and the difference between the two, for each of the three models

Differences between precipitation change in these experiments are shown for each model in Fig. 11. The overall pattern of change is quite similar between the two experiments in HadGEM2 and CCSM4, with correlation coefficients of 0.76 and 0.80 respectively, suggesting that present-day SST biases do not have an overwhelming influence on future projections in these models. However there are some regions that are clearly affected by SST biases, particularly the Maritime Continent and tropical oceans in both models, and central America and southern India in CCSM4. In these regions, the coupled projections of the two models should be treated with caution, as they may be heavily influenced by SST biases. In CNRM-CM5 the difference between the experiments is relatively more pronounced, with a correlation coefficient between the two of 0.56. In several regions, uncertainty introduced by present-day SST biases is of similar magnitude to the climate change signal itself, partly due to the relatively weak pattern of precipitation change in CNRM-CM5. Again, the maritime continent stands out as a region where future precipitation projections are particularly compromised by SST biases.

It has recently been suggested that the influence of present-day SST biases on uncertainty in future regional climate change over land may be greater than the uncertainty introduced by inter-model disagreement on the future pattern of SST change (He and Soden 2016). That study used results from a single atmospheric model forced with a variety of SST boundary conditions, and the CMIP6 a4SST-4x\(CO2\) multi-model experiment will allow the result to be tested across a number of different models. From the 3-model sample in the present study, uncertainty in both present-day and future SST patterns both appear to contribute to uncertainty in future precipitation and circulation change, but it is not clear which factor is more important.

6 Summary and conclusions

The new timeslice experiments are able to recreate the coupled model responses to \(CO_2\) forcing in the three models examined here, and can therefore be used to separate out regional responses to the various aspects of SST warming and \(CO_2\) forcing. This can be performed for any region or variable of interest, and has been applied here to two cases where inter-model uncertainty is particularly large: tropical water cycle change, and European winter circulation change.

The balance of mechanisms acting on tropical precipitation varies by region and by model. SST pattern change is the dominant driver of precipitation change over the oceans in CCSM4 and HadGEM2, but weaker gradients of SST change in CNRM-CM5 lead to a weaker influence on precipitation change, and an overall more muted pattern of precipitation change in that model. This result suggests that the use of a common multi-model ensemble mean, and therefore smoother, pattern of SST change in the CMIP5 idealised amipFuture experiments could lead to an underestimate of the role of the pattern of SST anomalies as a source of regional precipitation change. In the present study, SST pattern change is also influential for precipitation change in a number of tropical land regions, suggesting that uncertainty in the pattern of future SST change contributes to uncertainty in regional tropical precipitation change over land as well as over ocean.

A notable result is the dominance of the plant physiological effect on water cycle responses in tropical forest regions (see also Boisier 2017). As well as directly reducing transpiration, this produces large changes in precipitation and soil moisture, and is responsible for much larger overall tropical forest hydrological cycle changes in HadGEM2-ES and CCSM4, which include the plant effect, than in CNRM-CM5, which does not. This suggests that the CMIP5 ensemble mean, which includes a number of GCMs without the plant effect, may be underestimating the size of future water cycle changes in tropical forest regions. This is an example of how these timeslice experiments can be used to constrain coupled model uncertainty, by linking regional climate change to underlying model processes.

Over Europe and the North Atlantic, a uniform SST increase produces a northward shift of the stormtrack in all three models, although the exact pattern of change differs between the three. However this poleward response is not reflected in the total pattern of PMSL change in the models, largely because there is a large circulation response to SST pattern change in each model. These circulation responses vary greatly between the three models, and could be caused by variation in the local relative cooling of the North Atlantic associated with a slow-down of the AMOC (Woollings 2012), or with teleconnections from tropical SST and rainfall anomalies (Deser and Phillips 2009; Fereday et al. 2017). In either case, SST pattern change appears to be the main reason why North Atlantic circulation change does not produce a simple polewards shift of the stormtrack.

In CCSM4 and HadGEM2, model SST biases appear to be influential in setting the pattern of future precipitation change in some tropical regions, such as the Maritime Continent, but they do not substantially affect the overall pattern of change. In CNRM-CM5, SST biases may influence the pattern of projected precipitation change to a greater extent, partly due to the generally weaker response of regional precipitation change to warming in this model. Model biases in other quantities, such as a common dry precipitation bias over the Amazon region (Joetzjer et al. 2013), are also likely to affect the fidelity of future projections in all models, but are not directly addressed by these experiments. However, improved understanding of the processes that drive regional climate change should help to clarify potential links between present-day biases and future projections, and potentially narrow projection uncertainty.

The results presented in this study will be followed up with analysis of a larger model ensemble of CMIP6 runs, and this should allow inter-model uncertainty in regional climate change to be much better understood, and potentially constrained. The final CMIP6 experimental design (Webb et al. 2016) will also allow the influence of sea-ice change to be isolated and examined.

Notes

Acknowledgements

We thank Sophie Tyteca for producing the CNRM-CM5 piSST simulations, and Noah Diffenbaugh and the Stanford Research Computing Center for providing computational resources and support that have contributed to these research results. We also thank Gill Martin and three anonymous reviewers for useful comments on the manuscript. Rob Chadwick was supported by the Newton Fund through the Met Office Climate Science for Service Partnership Brazil (CSSP Brazil). Chris Skinner was supported by the Turner Postdoctoral Fellowship through the University of Michigan.

References

  1. Ackerley D, Dommenget D (2016) Atmosphere-only GCM simulations with prescribed land surface temperatures. Geosci Model Dev. doi: 10.5194/gmd-2016-6
  2. Anderson B, Lintner B, Langenbrunner B, Neelin J, Hawkins E, Syktus J (2015) Sensitivity of terrestrial precipitation trends to the structural evolution of sea surface temperatures. Geophys Res Lett 42. doi: 10.1002/2014GL062593
  3. Andrews T, Forster PM, Gregory JM (2009) A surface energy perspective on climate change. J Clim 22(10):2557–2570CrossRefGoogle Scholar
  4. Barsugli J, Battisti D (1998) The basic effects of atmosphere–ocean thermal coupling on midlatitude variability. J Atmos Sci 55(4):477–493CrossRefGoogle Scholar
  5. Bayr T, Dommenget D (2013) The tropospheric land–sea warming contrast as the driver of tropical sea level pressure changes. J Clim 26(4):1387–1402CrossRefGoogle Scholar
  6. Bengtssonn L, Roeckner E (2006) Storm tracks and climate change. J Clim 19:3518–3543CrossRefGoogle Scholar
  7. Betts R, Boucher O, Collins M, Cox P, Falloon P, Gedney N, Hemming D, Huntingford C, Jones C, Sexton D, Webb M (2007) Projected increase in continental runoff due to plant responses to increasing carbon dioxide. Nature 448:1037–1042CrossRefGoogle Scholar
  8. Betts R, Cox P, Collins M, Harris P, Huntingford C, Jones C (2008) The role of ecosystem–atmosphere interactions in simulated Amazonian precipitation decrease and forest dieback under global climate warming. Theor Appl Climatol 78:157–175Google Scholar
  9. Biasutti M (2013) Forced Sahel rainfall trends in the CMIP5 archive. J Geophys Res 118(4):1613–1623Google Scholar
  10. Boisier JP (2017) Robust CO2-physiological effect on modelled rainforest water fluxes and its contribution to the Amazonian drying (In Prep.) Google Scholar
  11. Bony S, Bellon G, Klocke D, Sherwood S, Fermepin S, Denvil S (2013) Robust direct effect of carbon dioxide on tropical circulation and regional precipitation. Nat Geosci 6(6):447–451CrossRefGoogle Scholar
  12. Brown J, Matear R, Brown J, Katzfey J (2015) Precipitation projections in the tropical Pacific are sensitive to different types of SST bias adjustment. Geophys Res Lett. doi: 10.1002/2015GL066184
  13. Cao L, Bala G, Caldeira K (2012) Climate response to changes in atmospheric carbon dioxide and solar irradiance on the time scale of days to weeks. Environ Res Lett 7(3):034,015CrossRefGoogle Scholar
  14. Cash B, Schneider E, Bengtsonn L (2005) Origin of regional climate differences: role of boundary conditions and model formulation in two GCMs. Clim Dyn 25:709–723CrossRefGoogle Scholar
  15. Cattiaux J, Douville H, Peings Y (2013) European temperatures in CMIP5: origins of present-day biases and future uncertainties. Clim Dyn 41:2889–2907CrossRefGoogle Scholar
  16. Catto J, Shaffrey L, Hodges K (2011) Northern hemisphere extratropical cyclones in a warming climate in the HiGEM high-resolution climate model. J Clim 24:5336–5352CrossRefGoogle Scholar
  17. Chadwick R (2016) Which aspects of CO\(_2\) forcing and SST warming cause most uncertainty in projections of tropical rainfall change over land and ocean? J Clim. doi: 10.1175/JCLI-D-15-0777.1
  18. Chadwick R, Good P (2013) Understanding non-linear tropical precipitation responses to \(CO_2\) forcing. Geophys Res Lett 40. doi: 10.1002/grl.50932
  19. Chadwick R, Boutle I, Martin G (2013) Spatial patterns of precipitation change in CMIP5: why the rich do not get richer in the tropics. J Clim 26:3803–3822CrossRefGoogle Scholar
  20. Chadwick R, Good P, Andrews T, Martin G (2014) Surface warming patterns drive tropical rainfall pattern responses to \(CO_2\) forcing on all timescales. Geophys Res Lett 41. doi: 10.1002/2013GL058504
  21. Chadwick R, Good P, Martin G, Rowell D (2015) Large rainfall changes consistently projected over substantial areas of tropical land. Nat Clim Chang. doi: 10.1038/nclimate2805
  22. Chen X, Zhou T (2015) Distinct effects of global mean warming and regional sea surface warming pattern on projected uncertainty in the south Asian summer monsoon. Geophys Res Lett. doi: 10.1002/2015GL066384
  23. Collins M, Knutti R, Arblaster J, Dufresne JL, Fichefet T, Friedlingstein P, Gao X, Gutowski WJ, Johns T, Krinner G, Shongwe M, Tebaldi C, Weaver A, Wehner M (2013) Climate change 2013: the physical science basis. Contribution of working group 1 to the fifth assessment report of the Intergovernmental Panel on Climate Change. Cambridge Univ. Press, pp 1054–1099Google Scholar
  24. Deser C, Phillips A (2009) Atmospheric circulation trends, 1950–2000: the relative roles of sea surface temperature forcing and direct atmospheric radiative forcing. J Clim 22:396–413CrossRefGoogle Scholar
  25. Dong B, Gregory JM, Sutton RT (2009) Understanding land–sea warming contrast in response to increasing greenhouse gases. Part I: Transient adjustment. J Clim 22(11):3079–3097CrossRefGoogle Scholar
  26. Dong B, Sutton RT (2015) Dominant role of greenhouse-gas forcing in the recovery of Sahel rainfall. Nat Clim Chang 5:757–760. doi: 10.1038/nclimate2664 CrossRefGoogle Scholar
  27. Douville H (2005) Limitations of time-slice experiments for predicting regional climate change over south Asia. Clim Dyn 24:373–391CrossRefGoogle Scholar
  28. Fasullo J (2012) A mechanism for landocean contrasts in global monsoon trends in a warming climate. Clim Dyn 39:1137–1147CrossRefGoogle Scholar
  29. Fereday D, Chadwick R, Knight J, Scaife A (2017) Climate dynamics is the largest source of uncertainty in future UK/European rainfall. J Clim (In Prep.) Google Scholar
  30. Folland C, Sexton D, Karoly D, Johnson C, Rowell D, Parker D (1998) Influences of anthropogenic and oceanic forcing on recent climate change. Geophys Res Lett 25(3):353–356CrossRefGoogle Scholar
  31. Gaetani M, Flamant C, Bastin S, Janicot S, Lavaysse C, Hourdin F, Braconnot P, Bony S (2016) West African monsoon dynamics and precipitation: the competition between global SST warming and \(CO_2\) increase in CMIP5 idealised simulations. Clim Dyn. doi: 10.1007/s00382-016-3146-z
  32. Gastineau G, Li L, Le Treut H (2009) The Hadley and Walker circulation changes in global warming conditions described by idealized atmospheric simulations. J Clim 22(14):3993–4013CrossRefGoogle Scholar
  33. Giannini A (2010) Mechanisms of climate change in the semiarid African Sahel: the local view. J Clim 23:743756CrossRefGoogle Scholar
  34. Giannini A, Salack S, Lodoun T, Ali A, Gaye A, Ndiaye O (2013) A unifying view of climate change in the Sahel linking intra-seasonal, interannual and longer time scales . Environ Res Lett 8. doi: 10.1088/1748-9326/8/2/024010
  35. Good P, Booth B, Chadwick R, Hawkins E, Jonko A, Lowe J (2016) Large differences in regional precipitation change between a first and second 2K of global warming. Nat Commun 7. doi: 10.1038/ncomms13667
  36. Grise K, Polvani L (2014) The response of midlatitude jets to increased CO\(_{2}\): distinguishing the roles of sea surface temperature and direct radiative forcing. Geophys Res Lett. doi: 10.1002/2014GL061638
  37. Harris P, Huntingford C, Cox P (2008) Amazon Basin climate under global warming: the role of the sea surface temperature. Philos Trans R Soc B 363:1753–1759CrossRefGoogle Scholar
  38. Harvey B, Shaffrey L, Woollings T (2014) Equator-to-pole temperature differences and the extra-tropical storm track responses of the CMIP5 climate models. Clim Dyn 43:1171–1182CrossRefGoogle Scholar
  39. Harvey B, Shaffrey L, Woollings T (2015) Deconstructing the climate change response of the Northern Hemisphere wintertime stormtracks. Clim Dyn 45:2847–2860CrossRefGoogle Scholar
  40. Hawkins E, Sutton R (2011) The potential to narrow uncertainty in projections of regional precipitation change. Clim Dyn 37(1):407–418CrossRefGoogle Scholar
  41. Held IM, Delworth T, Lu J, Findell K, Knutson T (2005) Simulation of Sahel drought in the 20th and 21st centuries. Proc Natl Acad Sci USA 102. doi: 10.1073/pnas.0509057102
  42. Held IM, Soden BJ (2006) Robust responses of the hydrological cycle to global warming. J Clim 19(21):5686–5699CrossRefGoogle Scholar
  43. He J, Soden B (2015a) Anthropogenic weakening of the tropical circulation: the relative roles of direct CO\(_{2}\) forcing and sea surface temperature change. J Clim. doi: 10.1175/JCLI-D-15-0205.1
  44. He J, Soden B (2015b) Does the lack of coupling in SST-forced atmosphere-only models limit their usefulness for climate change studies? J Clim. doi: 10.1175/JCLI-D-14-00597.1
  45. He J, Soden B (2016) The impact of SST biases on projections of anthropogenic climate change: a greater role for atmosphere-only models? Geophys Res Lett. doi: 10.1002/2016GL069803
  46. He J, Soden B, Kirtman B (2014) The robustness of the atmospheric circulation and precipitation response to future anthropogenic surface warming. Geophys Res Lett. doi: 10.1002/2014GL059435
  47. Huang P, Ying J (2015) A multimodel ensemble pattern regression method to correct the tropical Pacific SST change patterns under global warming. J Clim 28:4706–4723CrossRefGoogle Scholar
  48. Joetzjer E, Douville H, Delire C (2013) Present-day and future Amazonian precipitation in global climate models: CMIP5 versus CMIP3. Clim Dyn. doi: 10.1007/s00382-012-1644-1
  49. Kamae Y, Ogura T, Watanabe M, Xie SP, Ueda H (2016) Robust cloud feedback over tropical land in a warming climate. J Geophys Res. doi: 10.1002/2015JD024525
  50. Kent C, Chadwick R, Rowell D (2015) Understanding uncertainties in future projections of seasonal tropical precipitation. J Clim 28:4390–4413CrossRefGoogle Scholar
  51. Li W, Fu R, Dickinson R (2006) Rainfall and its seasonality over the Amazon in the 21st century as assessed by the coupled models for the IPCC AR4. J Geophys Res 111. doi: 10.1029/2005JD006355
  52. Long SM, Xie SP (2015) Intermodel variations in projected precipitation change over the North Atlantic: sea surface temperature effect. Geophys Res Lett 42:4158–4165. doi: 10.1002/2015GL063852 CrossRefGoogle Scholar
  53. Long SM, Xie SP, Liu W (2016) Uncertainty in tropical rainfall projections: atmospheric circulation effect and the ocean coupling. J Clim 29:2671–2687CrossRefGoogle Scholar
  54. Ma J, Xie SP (2013) Regional patterns of sea surface temperature change: a source of uncertainty in future projections of precipitation and atmospheric circulation. J Clim 26(8):2482–2501CrossRefGoogle Scholar
  55. Ma J, Xie SP, Kosaka Y (2012) Mechanisms for tropical tropospheric circulation change in response to global warming. J Clim 25(8):2979–2994CrossRefGoogle Scholar
  56. Manabe S, Bryan K, Spelman M (1990) Transient response of a global ocean–atmosphere model to a doubling of atmospheric carbon dioxide. J Phys Oceanogr 20:722–749CrossRefGoogle Scholar
  57. Martin GM, Bellouin N, Collins WJ, Culverwell ID, Halloran PR, Hardiman SC, Hinton TJ, Jones CD, McDonald RE, McLaren AJ, O’Connor FM, Roberts MJ, Rodriguez JM, Woodward S, Best MJ, Brooks ME, Brown AR, Butchart N, Dearden C, Derbyshire SH, Dharssi I, Doutriaux-Boucher M, Edwards JM, Falloon PD, Gedney N, Gray LJ, Hewitt HT, Hobson M, Huddleston MR, Hughes J, Ineson S, Ingram WJ, James PM, Johns TC, Johnson CE, Jones A, Jones CP, Joshi MM, Keen AB, Liddicoat S, Lock AP, Maidens AV, Manners JC, Milton SF, Rae JGL, Ridley JK, Sellar A, Senior CA, Totterdell IJ, Verhoef A, Vidale PL, Wiltshire A (2011) The HadGEM2 family of Met Office unified model climate configurations. Geosci Model Dev 4:723–757CrossRefGoogle Scholar
  58. McSweeney CF, Jones RG (2013) No consensus on consensus: the challenge of finding a universal approach to measuring and mapping ensemble consistency in GCM projections. Clim Chang 119:617–629CrossRefGoogle Scholar
  59. Meehl G, Washington W, Arblaster J, Hu A, Teng H, Tebaldi C, Sanderson B, Lamarque JF, Conley A, Strand W, White J (2012) Climate system response to external forcings and climate change projections in CCSM4. J Clim 25:3661–3683CrossRefGoogle Scholar
  60. Merlis T (2015) Direct weakening of tropical circulations from masked CO\(_2\) radiative forcing. Proc Nat Acad Sci 112(43):13167–13171CrossRefGoogle Scholar
  61. Oueslati B, Bony S, Risi C, Dufresne JL (2016) Interpreting the intermodel spread in regional precipitation projections in the tropics: role of surface evaporation and cloud radiative effects. Clim Dyn. doi: 10.1007/s00382-016-2998-6
  62. Richardson T, Forster P, Andrews T, Parker D (2016) Understanding the rapid precipitation response to CO\(_{2}\) and aerosol forcing on a regional scale. J Clim. doi: 10.1175/JCLI-D-15-0174.1
  63. Scaife A, Spangehl T, Fereday D, Cubasch U, Langematz U, Akiyoshi H, Bekki S, Braesicke P, Butchart N, Chipperfield M, Gettelman A, Hardiman S, Michou M, Rozanov E, Shepherd T (2012) Climate change projections and stratospheretroposphere interaction. Clim Dyn 38:2089–2097CrossRefGoogle Scholar
  64. Schneider T, O’Gorman PA, Levine X (2010) Water vapor and the dynamics of climate changes. Rev Geophys 48. doi: 10.1029/2009RG000302
  65. Seager R, Naik N, Vecchi GA (2010) Thermodynamic and dynamic mechanisms for large-scale changes in the hydrological cycle in response to global warming. J Clim 23(17):4651–4668CrossRefGoogle Scholar
  66. Shaw T, Voigt A (2015) Tug of war on summertime circulation between radiative forcing and sea surface warming. Nat Geosci. doi: 10.1038/NGEO2449
  67. Shepherd T (2014) Atmospheric circulation as a source of uncertainty in climate change projections. Nat Geosci 7:703–708CrossRefGoogle Scholar
  68. Shin S, Sardeshmukh P (2011) Critical influence of the pattern of tropical ocean warming on remote climate trends. Clim Dyn 36:1577–1591CrossRefGoogle Scholar
  69. Skinner C, Ashfaq M, Diffenbaugh N (2012) Influence of twenty-first century atmospheric and sea surface temperature forcing on West African climate. J Clim 25:527–541CrossRefGoogle Scholar
  70. Skinner C, Poulsen C, Chadwick R, Diffenbaugh N, Fiorella R (2016) The role of plant CO2 physiological forcing in shaping future daily-scale precipitation. J Clim. doi: 10.1175/JCLI-D-16-0603.1
  71. Sugi M, Yoshimura J (2004) A mechanism of tropical precipitation change due to CO\(_{2}\) increase. J Clim 17(1):238–243CrossRefGoogle Scholar
  72. Taylor KE, Stouffer RJ, Meehl GA (2012) An overview of CMIP5 and the experiment design. Bull Am Meteorol Soc 93(4):485–498CrossRefGoogle Scholar
  73. Thorpe L, Andrews T (2014) The physical drivers of historical and 21st century global precipitation changes. Environ Res Lett 9. doi: 10.1088/1748-9326/9/6/064024
  74. Voigt A, Shaw T (2015) Circulation response to warming shaped by radiative changes of clouds and water vapour. Nat Geosci 8:102–106CrossRefGoogle Scholar
  75. Voldoire A, Sanchez-Gomez E, Mlia DS, Decharme B, Cassou C, Snsi S, Valcke S, Beau I, Alias A, Chevallier M, Dqu M, Deshayes J, Douville H, Fernandez E, Madec G, Maisonnave E, Moine MP, Planton S, Saint-Martin D, Szopa S, Tyteca S, Alkama R, Belamari S, Braun A, Coquart L, Chauvin F (2011) The CNRM-CM5.1 global climate model: description and basic evaluation. Clim Dyn. doi: 10.1007/s00382-011-1259-y
  76. Webb M, Andrews T, Bodas-Salcedo A, Bony S, Bretherton C, Chadwick R, Chepfer H, Douville H, Good P, Kay J, Klein S, Marchand R, Medeiros B, Siebesma A, Skinner C, Stevens B, Tselioudis G, Tsushima Y, Watanabe M (2016) The Cloud Feedback Model Intercomparison Project (CFMIP) contribution to CMIP6. Geosci Model Dev. doi: 10.5194/gmd-2016-70
  77. Woollings T, Gregory J, Pinto J, Reyers M, Brayshaw D (2012) Response of the north Atlantic storm track to climate change shaped by ocean–atmosphere coupling. Nat Geosci 5. doi: 10.1038/NGEO1438
  78. Wu R, Kirtmann B (2004) Impacts of the Indian ocean on the Indian summer monsoon–ENSO relationship. J Clim 17:3037–3054CrossRefGoogle Scholar
  79. Xie SP, Deser C, Vecchi GA, Ma J, Teng H, Wittenberg AT (2010) Global warming pattern formation: sea surface temperature and rainfall. J Clim 23(4):966–986CrossRefGoogle Scholar
  80. Xie SP, Deser C, Vecchi GA, Collins M, Delworth T, Hall A, Hawkins E, Johnson N, Cassou C, Giannini A, Watanabe M (2015) Towards predictive understanding of regional climate change. Nat Clim Chang 5:921–930. doi: 10.1038/nclimate2689 CrossRefGoogle Scholar
  81. Zappa G, Shaffrey L, Hodges K, Sansom P, Stephenson D (2013) A multimodel assessment of future projections of North Atlantic and European extratropical cyclones in the CMIP5 climate models. J Clim 26:5846–5862CrossRefGoogle Scholar
  82. Zhang L, Li T (2016) Relative roles of differential SST warming, uniform SST warming and land surface warming in determining the Walker circulation changes under global warming. Clim Dyn. doi: 10.1007/s00382-016-3123-6

Copyright information

© © Crown Copyright as represented by the Met Office 2017

Authors and Affiliations

  1. 1.Met OfficeExeterUK

Personalised recommendations