Climate models suffer from long-standing biases, including the double intertropical convergence zone (ITCZ) problem and the excessive westward extension of the equatorial Pacific cold tongue. An atmospheric general circulation model is used to investigate how model biases in the mean state affect the projection of tropical climate change. The model is forced with a pattern of sea surface temperature (SST) increase derived from a coupled simulation of global warming but uses an SST climatology derived from either observations or a coupled historical simulation. The comparison of the experiments reveals that the climatological biases have important impacts on projected changes in the tropics. Specifically, during February–April when the climatological ITCZ displaces spuriously into the Southern Hemisphere, the model overestimates (underestimates) the projected rainfall increase in the warmer climate south (north) of the equator over the eastern Pacific. Furthermore, the global warming–induced Walker circulation slowdown is biased weak in the projection using coupled model climatology, suggesting that the projection of the reduced equatorial Pacific trade winds may also be underestimated. This is related to the bias that the climatological Walker circulation is too weak in the model, which is in turn due to a too-weak mean SST gradient in the zonal direction. The results highlight the importance of improving the climatological simulation for more reliable projections of regional climate change.
Precipitation changes in a warmer climate have major impacts on water security, with implications for many other sectors, including agriculture production, ecosystems, and human health. Tropical rainfall is projected to undergo significant changes in both magnitude and spatial pattern in response to global warming (Collins et al. 2013; Christensen et al. 2013). There are still large uncertainties in rainfall projections for tropical regions, as reflected in the large range of rainfall projections among different models from both the phase 3 of the Coupled Model Intercomparison Project (CMIP3) archive (Meehl et al. 2007) and the CMIP5 archive (Taylor et al. 2012). In many regions even the sign of change is uncertain (van Oldenborgh et al. 2013). These large uncertainties come from three sources: model uncertainty, scenario uncertainty, and internal variability of climate. The model uncertainty refers to the fact that each general circulation model (GCM) projects somewhat different future climate changes in response to the same radiative forcing; the scenario uncertainty arises from the uncertain future changes in anthropogenic forcing; and internal climate fluctuations introduce errors in estimating anthropogenic changes over a decade or more. Though GCMs can give us some estimates of rainfall projection uncertainties, such an approach is limited in use because of common model biases.
Climatological sea surface temperature (SST) is important for future rainfall projections. The ocean warming displays pronounced spatial variations in response to anthropogenic radiative forcing (Xie et al. 2010). For example, many models project an SST warming peak on the equator in the tropical Pacific (Liu et al. 2005) as a result of processes such as reduced evaporative damping in the climatological cold tongue [Fig. 14.14 of Christensen et al. (2013)]. Changes in tropical convection are strongly influenced by SST spatial structure (Xie et al. 2010; Chadwick et al. 2013; Chung et al. 2014; Ma and Xie 2013; Huang et al. 2013). There are two mechanisms for tropical rainfall change: the wet-get-wetter mechanism (rainfall increases in presently rainy regions) (Neelin et al. 2003; Held and Soden 2006; Chou et al. 2009) and the warmer-get-wetter mechanism (rainfall increases where the local SST increase exceeds the tropical mean warming) (Xie et al. 2010; Johnson and Xie 2010; Sobel and Camargo 2011; Chadwick et al. 2013). Under global warming, the atmospheric water vapor shows a robust increase as relative humidity remains relatively unchanged. The resultant intensification of the vertical moisture gradient, advected by the mean vertical motion, causes rainfall to increase over the wet regions, and vice versa (wet get wetter). The wet-get-wetter mechanism is related to increases in atmospheric moisture and associated with a slowdown of the overturning circulation, especially the Walker circulation in the tropical Pacific (Held and Soden 2006; Vecchi et al. 2006). The upper troposphere is projected to warm nearly uniformly in the horizontal because of fast equatorial wave actions (Johnson and Xie 2010). As a result, tropical convective instability change is determined by low-level moisture change, which largely follows the SST warming pattern (warmer get wetter; Xie et al. 2010; Chadwick et al. 2014).
State-of-the-art GCMs used for climate prediction and projection suffer from large errors in simulating tropical climate, including the double intertropical convergence zone (ITCZ) and excessive equatorial Pacific cold tongue (e.g., Mechoso et al. 1995; de Szoeke and Xie 2008; Richter and Xie 2008; Li and Xie 2012, 2014; Grose et al. 2014a) as well as insufficient trade winds in the equatorial eastern Pacific (Xiang et al. 2014). Conceivably, these errors affect future climate projections. The effect of the mean biases on projected future changes has not been quantified. Here we investigate the extent to which tropical climate projections depend on SST climatology.
The present study examines the sensitivity of tropical climate projections to SST climatologies under global warming and investigates the underpinning dynamical mechanisms. We impose a global SST warming pattern in the Community Atmosphere Model, version 4 (CAM4), that corresponds to the change in the late twenty-first century. We conduct two sets of experiments that differ only in their present-day climatological mean states—one based on observations and one based on the coupled model historical simulations.
2. Datasets and methods
CAM4 is the seventh-generation atmospheric general circulation model (AGCM) developed with significant community collaboration at the National Center for Atmospheric Research (NCAR). It is the atmospheric component of the Community Climate System Model, version 4 (CCSM4; Gent et al. 2011). A comprehensive description of CAM4 can be found in Neale et al. (2011). The model uses a finite-volume core (Lin 2004). We choose a 1.9° × 2.5° grid (f19_f19) in the horizontal and 26 sigma levels in the vertical. Table 1 summarizes the experiments conducted.
We employ the observed SST climatology from the National Oceanic and Atmospheric Administration (NOAA) monthly optimum interpolation sea surface temperature version 2 (OISST.v2) data for the years 1982–2008 (Reynolds et al. 2002). The CCSM4 SST climatology is derived as the 6-member ensemble mean of the historical run for the years 1979–2005. We use these two climatologies to drive the CAM4 for comparison (Fig. 1; referred to as Obs and Hist in figures). Two pairs of integrations are performed, with the following SST boundary conditions: 1) observed and historical climatologies (present pair in figures); 2) similar to the first pair but with an SST warming pattern added (future pair in figures). The SST warming pattern is derived from the CCSM4 representative concentration pathway (RCP) 8.5 run as 2081–2100 mean minus 1986–2005 (historical run) mean (Fig. 2), as in Zhou et al. (2014). We use the difference between the present and future pairs to diagnose tropical rainfall projections under different basic states. We run each experiment (annual cycle only, no interannual variability) for 30 years, and the last 29 years are used for analysis; the first year is considered as model spinup and not included. The radiative forcing (e.g., CO2) levels are kept constant at the observed level (379 ppm) throughout all the experiments. We choose the above SST warming pattern because of the model consistency since CAM4 is the atmospheric component of CCSM4. This SST warming pattern is similar to the multimodel ensemble mean warming pattern in Fig. 14.14 of Christensen et al. (2013).
As most of the state-of-art GCMs suffer from climatological biases, the SST warming pattern projected by these GCMs may be influenced by the biases. The model bias effect on the SST warming pattern is not investigated here but is an important issue that requires community effort through improving models and further innovative research (Huang and Ying 2015). Models project somewhat different SST warming patterns, but the dependency of the model bias effect on the SST warming pattern is not investigated here either.
3. Double ITCZ bias
We investigate the sensitivity of tropical climate projections to different SST climatologies. Figure 1 compares the annual mean basic states. Compared to observations (Fig. 1a), there are many regional differences in the CCSM4 historical climatology: the Indo-Pacific warm pool is displaced slightly southward; SST is too cold over the Maritime Continent, South China Sea, and Bay of Bengal and too warm in the other parts; the warm bands under the ITCZ and the South Pacific convergence zone (SPCZ) extend far too eastward; there are warm SST biases (>0.4 K) in the tropical southeast Pacific and the tropical southeast Atlantic; and SST is too cool in the equatorial Pacific cold tongue (over −0.3 K) and the tropical northwest Atlantic (over −0.6 K), with a weaker east–west SST gradient. The tropical mean SST is about −0.18 K compared to observations (Fig. 1c). As changes in tropical convection are strongly influenced by SST spatial structure (Xie et al. 2010; Chadwick et al. 2013; Ma and Xie 2013; Huang et al. 2013; Chung et al. 2014; Brown et al. 2015), the annual mean tropical rainfall pattern in the Hist run is also different from the Obs run (Fig. 3), notably with excessive rainfall south of the equator. Hereafter, the overbar and prime denote the climatological mean and global warming-induced change, respectively, and Δ denotes the difference between the Hist and the Obs runs.
Figures 3a, 3c, and 3e show annual mean rainfall climatology for the present climate. Corresponding to the cross-equatorial dipole of cold biases in the north and warm biases in the south, rainfall in the ITCZ is underestimated and rainfall in the tropical southeast Pacific and the southeast Atlantic are overestimated in the Hist run compared to the Obs run. In addition, rainfall in the Indian Ocean has a similar bias pattern to the ΔSST pattern, with overestimated (underestimated) rainfall over the southwestern (northeastern) basin. The annual mean climatologies of rainfall in the future warmer climate (not shown) show similar differences between Hist and Obs runs.
Figures 3b and 3d show the global warming–induced rainfall changes . The El Niño–like warming pattern is associated with a projected increase of convection on the equator (Figs. 3b,d) in CMIP3 (Ma and Xie 2013) and CMIP5 models (Chadwick et al. 2013; Christensen et al. 2013), following the warmer-get-wetter pattern (Xie et al. 2010; Grose et al. 2014b). But there are significant differences between rainfall changes in the Hist and the Obs runs. The rainfall change difference between Hist and Obs (Δ) is similar to their difference in climatological rainfall Δ, with much greater (less) rainfall increase south (north) of the equator in the Hist run than in the Obs run (Fig. 3f), following the wet-get-wetter mechanism, as the warming pattern is the same between the two pair of experiments. This indicates that the mean state has clear effects on the rainfall projections. The global warming response may amplify the mean-state biases.
Figure 4 shows latitude–time sections of zonal mean precipitation (color shaded; mm d−1) over the eastern Pacific (140°–80°W) for the present climatology and the global warming–induced changes, to illustrate the seasonal cycle of the double ITCZ bias effects. In the Obs run, the climatological rainband stays almost entirely north of the equator throughout the year (Fig. 4a). In Hist, the ITCZ moves south of the equator during austral summer and autumn (Fig. 4c), a bias called the double ITCZ problem. As a result, large excessive (deficient) rainfall occurs in the Southern (Northern) Hemisphere (SH; NH) during austral summer and autumn in the Hist run (Figs. 4c,e). In observations during February–April (FMA), the ocean–atmosphere system is nearly symmetric about the equator over the eastern Pacific, with a seasonal double ITCZ. In the Hist run, SST south of the equator exceeds that to the north, eliminating the ITCZ north of the equator. Thus, a spurious equatorial asymmetry develops during FMA in Hist over the eastern Pacific, with a southward displaced ITCZ and too-weak southeasterly (strong northeasterly) trade winds in the SH (NH) tropics (Li and Xie 2014).
The enhanced equatorial warming anchors an increase in annual precipitation near the equator (Figs. 4b,d). This pattern of rainfall change marches back and forth across the equator with the season (Huang et al. 2013). The bias in rainfall projection is most pronounced during January–May when the mean ITCZ is spuriously displaced south of the equator (Figs. 4e,f). During FMA, the projected rainfall change is too high south and too low north of the equator in the Hist run compared to the Obs run (Fig. 4f). Figure 5 examines the relationship of the bias in the rainfall projection with the climatological rainfall and SST biases over the eastern Pacific (140°–80°W) during FMA. The correlation is taken in time and space for the seasonal cycle within 10°S–10°N, a latitudinal band that pronounced precipitation change is confined to (Figs. 4e,f). Differences in FMA rainfall projection Δ are significantly correlated with the climatological rainfall differences in present and future runs and climatological SST differences ΔSST over the eastern Pacific (140°–80°W), with correlation coefficients in space at 0.68, 0.86, and 0.68, respectively. This supports the wet-get-wetter mechanism ( > 0 where > 0) in explaining the biases of rainfall change projection. It is worth noting that the correlation between Δ and is somewhat higher if the future climatological rainfall bias is used (Fig. 5b).
Figure 6 shows the spatial structure of FMA-mean global warming induced changes in rainfall and 850-mb wind velocity over the Pacific Ocean in Obs, Hist, and their differences. Consistent with the global warming–induced annual mean rainfall changes in Fig. 3, the enhanced equatorial warming anchors the projected precipitation increase on the equator, following the warmer-get-wetter pattern (Xie et al. 2010). Meanwhile, the Walker circulation weakens (westerly anomalies in Figs. 6a,b) in a warmer climate. The global warming–induced rainfall changes north (south) of the equator is underestimated (overestimated) over the eastern Pacific in the Hist run compared to the Obs run; the ratio of in the Hist run to that in the Obs run is 0.19 north (0°–7.5°N, 140°–80°W) and 1.27 south (0°–7.5°S, 140°–80°W) of the equator. This is consistent with the wet-get-wetter pattern and illustrates that errors in SST and precipitation climatologies would bias the projection of rainfall change.
4. Walker circulation
Held and Soden (2006) suggested a general weakening of the tropical circulation by noting a slower increase in global mean rainfall than in specific humidity. The observed Walker circulation weakened over the twentieth century (Vecchi et al. 2006) and over the last six decades (Tokinaga et al. 2012). In CMIP3 models, the Walker circulation weakens while changes in the Hadley cells are strongly affected by the SST pattern and variable among models (Vecchi and Soden 2007; Gastineau et al. 2009; Ma et al. 2012). The projected slowdown of the Walker circulation deepens the eastern Pacific thermocline, creating a thermocline feedback (Vecchi and Soden 2007).
Figures 7a, 7c, and 7e show 300-mb annual mean velocity potential climatology for the present climate. The 300-mb velocity potential displays a significant divergence over the Maritime Continent and convergence over the equatorial eastern Pacific and Atlantic (Figs. 7a,c), which represents a typical Walker circulation. Corresponding to a too-weak east–west gradient with warm SST biases over the tropical southeast Pacific and the tropical southeast Atlantic (Fig. 1), the climatological Walker circulation is underestimated in the Hist run compared to the Obs run (Fig. 7e). Figures 7b, 7d, and 7f show 300-mb annual mean velocity potential changes Φ′ induced by the same warming pattern in the Obs and Hist runs and their differences. Both the Obs and Hist runs show significant changes under the El Niño–like warming, including the Walker circulation slowdown and the eastward shift of atmospheric convection from the Indonesian Maritime Continent to the central tropical Pacific (Figs. 7b,d). This may be due to the enhanced mean SST warming in the eastern equatorial Pacific, which reduces the zonal SST gradient and lowers the barrier to deep convection in the eastern basin. We define the difference in velocity potential between the eastern Pacific (10°N–20°S, 130°–70°W) and the Maritime Continent (20°N–20°S, 90°–160°E,) as the intensity of the Walker circulation. Our results are not very sensitive to the choice of definition. The Walker circulation change is stronger in the Obs run (6.63 × ) than in the Hist run (5.62 × ; Fig. 7f). We note that the ratio of change to the mean in the Walker circulation intensity is similar between the Obs and Hist runs, at 0.28 and 0.31, respectively. Thus, the stronger Walker circulation response to global warming in the Obs run is due partly to the stronger circulation in the mean. This is broadly consistent with the theory of Held and Soden (2006), which predicts the ratio of change based on the muted global mean precipitation response to global warming.
Figure 8 shows the spatial structure of annual mean changes in rainfall and 850-mb wind velocity induced by global warming over the tropical oceans in the Hist and Obs runs and their differences. Consistent with the Walker circulation slowdown in the 300-mb velocity potential changes (Fig. 7), the anomalous 850-mb winds are westerly over the tropical Pacific and easterly over the tropical Indian and Atlantic (Figs. 8a,b). Meanwhile, rainfall decreases over the Maritime Continent and increases over the western Indian Ocean and the central and eastern Pacific (Figs. 8a,b). These changes are consistent with the reduced surface warming around the Maritime Continent compared to the eastern equatorial Pacific and western equatorial Indian Ocean (Fig. 2), a pattern representing reduced zonal gradients of SST. In addition, The Walker circulation slows down much more in the Obs run than in the Hist run, with westerly winds over the western Indian Ocean and weak easterly winds over the Pacific and Atlantic Oceans in the Hist − Obs difference (Fig. 8c), consistent with the model bias effect on the 300-mb velocity potential change in Fig. 7f.
Figure 9 shows global warming–induced precipitation changes along the equator (5°N–5°S mean). Precipitation changes in the Hist run are weaker than in the Obs run (red line in Fig. 9) over the central and eastern equatorial Pacific Ocean, as the excessive equatorial Pacific cold tongue bias increases the barrier to the convective threshold. The model bias effect on circulation and precipitation are mutually consistent.
5. Summary and discussion
We have investigated the sensitivity of global warming–induced tropical climate changes to different SST climatologies. From 30-yr CAM4 simulations forced with the SST warming pattern derived from CCSM4 RCP 8.5 (2081–2100 mean) minus historical run (1986–2005 mean), we show that the projected rainfall change differs significantly between using the observed and historical simulated SST climatologies. The SST climatology in the Hist run features an excessive equatorial Pacific cold tongue, and cold (warm) biases north (south) of the equator in the tropical eastern Pacific. These biases are associated with the double ITCZ biases, with deficient (excessive) climatological rainfall in the tropical northeastern (southeastern) Pacific.
Under global warming, the enhanced equatorial warming anchors precipitation increase on the equator following the warmer-get-wetter pattern. The tropical rainfall projection bias (Hist − Obs difference) shows a prominent double-ITCZ-like pattern with the excessive (deficient) precipitation in the SH (NH) tropics. The rainfall projection bias is most prominent in the SH tropics during January–May when the climatological ITCZ spuriously displaces south of the equator. This indicates the importance of the climatological rainfall distribution, consistent with the wet-get-wetter mechanism (Huang et al. 2013). Such biases in the projected rainfall change may affect ENSO (Power et al. 2013; Cai et al. 2014) and its teleconnection (Zhou et al. 2014).
The Walker circulation weakens under global warming, owing partly to the muted hydrological cycle response and partly to the weakened zonal SST gradient over the Indo-Pacific. The Walker circulation slowdown is much more pronounced in the Obs run than in the Hist run. The climatological Walker circulation is stronger in the Obs run than in the Hist run. We suggest that this bias in the climatological Walker circulation explains the difference in the projected change.
The CMIP5 models are widely used for global climate predictions and projections (Taylor et al. 2012). There are still large errors in the simulation of the mean state in these models. Our results support a recommendation from a recent review (Xie et al. 2015) that for more accurate projections of climate change, it is important to improve the simulation of the present climate. Some of the biases (e.g., the double ITCZ and cold tongue problems) in tropical climate have persisted for several model generations despite the continued effort to reduce these biases. Recent studies suggest that in addition to the importance of local ocean–atmosphere feedbacks, the tropical biases might originate from extratropical errors (Hwang and Frierson 2013; Wang et al. 2014; Li and Xie 2014). The challenge is to translate such improved understanding to improved model performance, with important implications for future climate projection.
We are grateful for valuable comments and discussions on this work with Gen Li. This work was supported by the National Basic Research Program of China (2012CB955600), National Natural Science Foundation of China (NSFC)–Shandong Joint Fund for Marine Science Research Centers (U1406401), the U.S. National Science Foundation (NSF) and the National Oceanic and Atmospheric Administration (NOAA). The National Center for Atmospheric Research (NCAR) provided the CAM4 model. Computational resources for CAM4 runs were provided by the Computing Center of the Key Laboratory of Physical Oceanography, Chinese Ministry of Education.