Using two fully coupled ocean–atmosphere models—Climate Model version 2.1 (CM2.1), developed at the Geophysical Fluid Dynamics Laboratory, and Climate Model version 2.5 (CM2.5), a new high-resolution climate model based on CM2.1—the characteristics and sources of SST and precipitation biases associated with the Atlantic ITCZ have been investigated.
CM2.5 has an improved simulation of the annual mean and the annual cycle of the rainfall over the Sahel and northern South America, while CM2.1 shows excessive Sahel rainfall and lack of northern South America rainfall in boreal summer. This marked improvement in CM2.5 is due to not only high-resolved orography but also a significant reduction of biases in the seasonal meridional migration of the ITCZ. In particular, the seasonal northward migration of the ITCZ in boreal summer is coupled to the seasonal variation of SST and a subsurface doming of the thermocline in the northeastern tropical Atlantic, known as the Guinea Dome. Improvements in the ITCZ allow for better representation of the coupled processes that are important for an abrupt seasonally phase-locked decay of the interannual SST anomaly in the northern tropical Atlantic.
Nevertheless, the differences between CM2.5 and CM2.1 were not sufficient to reduce the warm SST biases in the eastern equatorial region and Angola–Benguela area. The weak bias of southerly winds along the southwestern African coast associated with the excessive southward migration bias of the ITCZ may be a key to improve the warm SST biases there.
Climate conditions in the tropical Atlantic have to be simulated well in climate models for accurate prediction of Atlantic Hurricane and drought (or flood) in the Sahel and South America (Emanuel 2005; Vecchi and Soden 2007; Hagos and Cook 2009). Also, the impacts of the Atlantic variations are not restricted to the Atlantic basin and can have far-reaching effects (Dommenget et al. 2006; Zhang and Delworth 2006; Zhang et al. 2007; Sutton and Hodson 2005; Kucharski et al. 2011). However, many coupled GCMs suffer from serious biases in the tropical Atlantic (Davey et al. 2002). In particular, almost all Coupled Model Intercomparison Project phase 3 (CMIP3) (Meehl et al. 2007) climate models for the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4) show warm SST biases in the eastern equatorial Atlantic and produce a zonal SST gradient along the Atlantic equator with opposite sign to observations (Richter and Xie 2008). The warm SST bias in the eastern equatorial Atlantic still remains in the new generation of climate model, the Community Climate System Motel, version 4 (CCSM4) (Grodsky et al. 2012; Muñoz et al. 2012). The lack of a cold tongue in the equatorial Atlantic will be one reason why many coupled GCMs fail to simulate and predict the Atlantic Niño (Stockdale et al. 2006), which is the most major climate mode in the tropical Atlantic (Zebiak 1993; Carton and Huang 1994). Richter and Xie (2008) showed that the origin of the biases is in the anomalously weak trade winds along the equator, which are associated with the ITCZ being displaced anomalously southward in boreal spring. Richter et al. (2011) suggests that the zonal surface wind errors along the equator were partially due to deficient precipitation over equatorial South America and excessive precipitation over equatorial Africa. Recently, the University of Tokyo Coupled Model (Tozuka et al. 2006) was able to successfully simulate the zonal gradient of the annual mean SST (Doi et al. 2010). Interestingly, the model’s ability to simulate this feature depends on the deep convection scheme used (Tozuka et al. 2011).
In the tropical Atlantic climate, two types of air–sea coupled process are important: the zonal Bjerknes positive feedback (Bjerknes 1969) and the meridional wind–evaporation–SST (WES) positive feedback (Xie 1999). The Bjerknes feedback is associated with a zonal gradient of SST on the equator: 1) a weak zonal gradient of SST is responsible for a weaker easterly wind on the equator, 2) the weaker easterly winds deepen (shoal) the thermocline and 3) warms (cools) the SST in the eastern (western) equatorial region. The outcome is further weakened easterly winds. The Bjerknes feedback develops the Atlantic Niño, characterized as a warm SST anomaly in the eastern equatorial Atlantic during boreal summer (Zebiak 1993; Carton and Huang 1994). The WES positive feedback is the interhemispheric two-way air–sea interaction between wind and SST and is associated with meridional migration of the ITCZ: 1) an anomalously northward migration of the ITCZ causes southwesterly (southeasterly) wind anomalies in the northern (southern) tropics leading to weaker trade winds; 2) this results in less (more) evaporation and thus suppressed (enhanced) latent heat loss from ocean, leading to warmer (colder) SST in the northern (southern) tropical Atlantic; and 3) the outcome is further northward migration of the ITCZ. The dominance of this mechanism in growth of the Atlantic meridional mode has been discussed in previous work (Carton et al. 1996; Chang et al. 1997; Xie 1999). The Atlantic meridional mode is characterized by the cross-equatorial meridional gradient of SST anomaly in the tropical Atlantic during boreal spring (Servain 1991; Xie and Carton 2004 for a recent review).
The Atlantic climate modes are closely linked with not only air–sea interactions between wind and SST but also subsurface oceanic conditions. In the tropical Atlantic, two thermocline domes associated with ocean upwelling are found: the Angola Dome in the southeastern tropical Atlantic and the Guinea Dome in the northeastern tropical Atlantic (Mazeika 1967). Interannual variations of the Angola Dome in the South Atlantic are strongly influenced by the Atlantic Niño (Doi et al. 2007), while the interannual variations of the Guinea Dome are linked with the Atlantic meridional mode (Doi et al. 2009, 2010). In particular, Doi et al. (2010) pointed out that the Guinea Dome could play a critical role on the seasonal phase locking of the interannual variations of the northern tropical Atlantic SST (Fig. 18 of Doi et al. 2010). Although most studies on the Atlantic climate have focused on atmospheric forcing or the sea surface temperature field, considering the ocean dynamical roles of upwelling and stratification variations in climate models is important for understanding tropical Atlantic climate modes (Doi 2009) and hurricane intensity (Lloyd and Vecchi 2011).
In this paper, we investigate how tropical Atlantic biases are improved by comparing a fully coupled ocean–atmosphere climate model with a new high-resolution climate model (including some changes to parameterizations, numerics, and a land model). Our manuscript shows that many aspects of the simulation in the tropical Atlantic are significantly improved in the new model—yet others persist. The paper is organized as follows. In section 2, differences between coarse- and high-resolution coupled climate models and a brief description of observational datasets are given. In section 3, the annual mean and the annual cycle of SST and rainfall in two models are explored. In the first half of section 3, we discuss the most severe biases in the eastern equatorial Atlantic, which still persist in the new high-resolution model. In the latter half, we focus on the meridional migration of the ITCZ in boreal summer and its coupled link with SST and the Guinea Dome in the northern tropical Atlantic because the new model significantly improved the simulation of rainfall in the Sahel and South America. In section 4, we explore the interannual variation of the two models and observations with particular focus on the seasonal phase locking of the interannual variation in the northern tropical Atlantic SST. The final section is used for summary and discussions.
2. Models and observational datasets
a. GFDL CM2.1
Detailed formulations of the Geophysical Fluid Dynamics Laboratory Climate Model version 2.1 (GFDL CM2.1) are described by Delworth et al. 2006, Gnanadesikan et al. 2006, Stouffer et al. 2006, and Wittenberg et al. 2006. CM2.1 was part of the CMIP3 model comparison, which contributed to the IPCC AR4, and CM2.1 has been shown to perform quite well for many global climate metrics (Knutson et al. 2006; Russell et al. 2006). CM2.1 is also the basis of experimental seasonal to decadal forecast systems at the GFDL (Zhang et al. 2007). The oceanic component is based on the Modular Ocean Model version 4 (MOM4) code (Griffies et al. 2005). The horizontal resolution is 1° longitudinal and 1° latitudinal with enhanced tropical resolution (⅓° within 10° of the equator). There are 50 vertical levels and the vertical grid spacing is a constant 10 m over the top 220 m. Isopycnal mixing of tracers and layer thickness is based on the formulation by Gent and McWilliams (1990), Griffies et al. (1998), and Griffies (1998). The mixed layer is represented by the K-profile parameterization (KPP) vertical mixing (Large et al. 1994). The shortwave penetration depends on prescribed spatiotemporally varying chlorophyll (Sweeney et al. 2005).
The atmospheric component is the Atmospheric Model version 2.1 (AM2.1) (GFDL Global Atmospheric Model Development Team) (GAMDT 2004), which consists of a finite volume dynamical core (Lin 2004) with 24 vertical levels, 2° latitude by 2.5° longitude grid spacing, K-profile planetary boundary layer scheme (Look et al. 2000), and relaxed Arakawa–Schubert convection (Moorthi and Suarez 1992). The land model is Land Model, version 2 (LM2), which includes soil sensible and latent heat storage, groundwater storage, and stomatal resistance (GAMDT 2004). The coupled simulation is initialized from observed climatological oceanic condition at year 1 (Delworth et al. 2006) and then integrated subject to 1990 values of trace gases, insolation, aerosols, and land cover. This radiative forcing yields a present-day control experiment. The atmosphere, ocean, land, and sea ice exchange fluxes every 2 h and no flux adjustments are employed. We have calculated monthly climatologies by averaging model monthly mean output for the 300 years. Then, we define anomaly fields as deviations from the monthly mean climatologies after removing the decadal variability using an 8-yr running mean filter.
b. GFDL CM2.5
GFDL Climate Model version 2.5 (CM2.5) is a new high-resolution model version that derives closely from GFDL CM2.1 (Delworth et al. 2012). The oceanic component of CM2.5 uses a 0.25° horizontal resolution of MOM4p1 in the tropics with the z* vertical coordinate (Griffies 2010; Griffies et al. 2011), which varies from 28 km near the tropics to 8 km in polar regions. Its oceanic component is similar to that of the CM2.4 model used in Farneti et al. (2010). It is coupled to a 50-km horizontal resolution atmosphere model with 32 vertical levels on a cubed-sphere grid (Lin 2004; Putman and Lin 2007). This formulation avoids the numerical problem arising from the convergence of meridians at the poles and allows grid boxes of roughly equal area over the globe. No flux adjustments are employed. The ocean model does not contain a parameterization for mesoscale eddy mixing. The advective scheme has been modified to yield substantially lower numerical diffusion, and a substantially smaller explicit viscosity than CM2.1 is employed. The land model is LM3 (Shevliakova et al. 2009), which represents snow and rain interception on vegetation, as well as water phase change in the soil and snowpack. CM2.5 is initialized and forced in a similar fashion to CM2.1 (Delworth et al. 2012, 2006); the oceanic initial condition is taken from the end of a 1-yr spin up from observed climatological conditions at rest and the atmospheric initial condition is taken from the end of a simulation with prescribed SST.
We used monthly mean output from a 280-yr simulation of CM2.5 with 1990 radiative forcing. The monthly averaged climatology and interannual anomaly are calculated in the same manner as in CM2.1. For reference, the atmospheric components of CM2.1 and CM2.5 forced by the observed SST in 1981–2000 (AM2.1 and AM2.5) are also used.
c. Observational datasets
We use two SST datasets to evaluate the models: the Extended Reconstructed SST version 3 (ERSSTv3b) (Smith et al. 2008) and, the Hadley Center SST (HadISST) (Rayner et al. 2003). For wind stress and surface enthalpy fluxes, we used the NCEP/NCAR reanalysis data (Kalnay et al. 1996) and ECMWF 40-year reanalysis data (ERA-40) (Simmons and Gibson 2000). Also, surface enthalpy flux data from the objectively analyzed air–sea fluxes (OAFlux) (Yu et al. 2006) are used. Monthly climatologies are calculated by averaging monthly observational data over 1960–2001, and then interannual anomalies are defined as deviations from the monthly mean climatologies after subtracting the 8-yr running mean values. Since the OAFlux project provides only latent and sensible heat fluxes in this epoch, the turbulent enthalpy fluxes of the OAFlux are combined with the radiative fluxes from the National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis data (OAFlux-N), and with the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40) reanalysis data (OAFlux-E). For precipitation data, Climate Prediction Center Merged Analysis of Precipitation dataset (CMAP) (Xie and Arkin 1997) and Global Precipitation Climatology Project dataset (GPCP) (Adler et al. 2003) are used for 1979–2001. The observed mixed layer depth is estimated from the monthly climatology of the World Ocean Atlas 2005 data (WOA05) (Locarnini et al. 2006) as the depth at which the potential density becomes larger than the surface density by 0.125 kg m−3, as used by Levitus (1982).
3. Annual mean and annual cycle
We begin by exploring the annual mean fields of the tropical Atlantic SST in observations and the two climate models (Fig. 1). The SST near the northern South American coast is colder by about 1°C in CM2.1 than the observed SST of ERSSTv3. This bias is reduced in CM2.5, likely due to the more reasonable simulation of the North Brazil Current and the eddy activity in this area. Also, CM2.1 has a cold SST bias of ~1.5°C over the northern tropical Atlantic and a warm bias of ~2.5°C over the eastern equatorial region. Unfortunately, these SST biases are not significantly improved in CM2.5. Although the warm SST bias in the Angola–Benguela area is reduced by ~0.5°C in CM2.5 relative to CM2.1, a warm SST bias of ~4.5°C persists there. Though we only show comparison to the ERSSTv3b, these SST biases are almost the same relative to the HadISST data.
The largest bias of annual mean SST in CM2.1 and CM2.5 is a warm SST bias over the eastern equatorial region, which is also present in almost all CMIP3 climate models, particularly in boreal summer. Therefore, we explore boreal spring wind biases to assess the sources of the warm SST bias. Richter and Xie (2008) and Richer et al. (2012) showed that the warm SST bias has been attributed to a bias toward weak easterly equatorial winds in boreal spring, which drives the excessively warm SST and deep thermocline in the eastern equatorial region and is amplified through the Bjerknes positive feedback (Bjerknes 1969). As shown in Table 1a, the easterly trade wind stress along the equator in CM2.1 is only 25% of the observed wind stress in boreal spring. This bias in boreal spring trade wind is partially reduced in CM2.5, but the simulated easterly is still only half of the observed wind stress. Associated with the warm SST bias and weak easterly equatorial wind, the mixed layer depth in the southeastern tropical Atlantic is deeper by 10 m in CM2.1 and CM2.5 than observations (figure not shown). Interestingly, the atmospheric components of CM2.1 and CM2.5 forced by observed SST can capture a reasonable strength of the observed easterly wind stress, suggesting that these weak wind biases arise from coupled processes. The southerly winds in the southeastern tropical Atlantic may also contribute to the warm SST bias because the southerly winds induce cold upwelled water along the West African coast in the South Hemisphere, which extends westward by advection and Rossby wave propagation and cools the eastern equatorial SST (Philander and Pacanowski 1981). The southerly winds along the West African coast are also much weaker in CM2.1 and CM2.5 relative to observation (Table 1b). The southerly wind stress in CM2.1 is only 20% of the observed wind stress. This bias is marginally reduced in CM2.5, yet captures only 30% of the observed wind stress. The weak bias of southerly winds that appeared in CM2.1 and CM2.5 is also ubiquitous in the IPCC AR4 CMIP3 climate models (Doi et al. 2010). The atmospheric components of two coupled models forced by observed SST already show the weak southerly winds along the West African coast at only 50% of those observed. Therefore, the improvement of the weak bias of southerly winds stress in AGCMs may be a good step toward reducing the warm SST bias in the southeastern tropical Atlantic, as we further discuss in section 5.
Although the most serious SST biases are not improved by the differences between CM2.1 and CM2.5, the annual mean precipitation field is substantially improved in CM2.5 relative to CM2.1. Over northern South America, CM2.1 shows a deficient rainfall relative to observations, while CM2.5 simulates a considerably more reasonable rainfall distribution (Fig. 2 and Table 1c). It is noteworthy that regions that become rainier in CM2.5 relative to CM2.1 are associated with steeper topography in high mountain regions (Fig. 2d). Therefore, we hypothesize that the higher resolution of CM2.5 allows for these orographic features (and their impacts) to be better captured, while the low resolution in CM2.1 results in smoother topography. Consistent with this hypothesis, the improvement in northern South American rainfall is also evident in atmosphere-only models: high-resolution AM2.5 with LM3 (or LM2) shows more rainfall over northern South America than the coarse-resolution AM2.1 (figure not shown).
Rainfall in the Congo basin is well simulated in both CM2.1 and CM2.5 (Table 1d). Given the results of Richter et al. (2011) and the suggestion of Wahl et al. (2011), we expected some improvement of the equatorial zonal winds and the zonal SST gradient from improving the precipitation pattern between northern Brazil and the Congo. The winds are, indeed, marginally improved in CM2.5. However, this is not sufficient to correct the mean-state SST bias. Note that CM2.5 shows an excessive precipitation bias associated with the marine Atlantic ITCZ within 50°–20°W, which is associated with the 0.5°C warmer mean SST in CM2.5 relative to CM2.1 (Fig. 1c and Fig. 2).
The annual cycle of rainfall over the Sahel and northern South America is shown in Fig. 3. In CM2.1, there is excessive rainfall (by 40%) in the Sahel, while rainfall over northern South America is deficient—particularly in boreal summer. Boreal summer rainfall over northern South America in CM2.1 is only 20% of that observed. Meanwhile, the annual cycle of rainfall over the Sahel and northern South America is drastically improved in CM2.5. This marked improvement in CM2.5 is associated with a significant reduction of some biases in the seasonal meridional migration of the ITCZ (Fig. 4). In CM2.1, the ITCZ shows an excessively large northward migration in boreal summer: the simulated ITCZ in CM2.1 is located around 3°S in boreal spring and moves northward to 12°N in boreal summer, while the observed ITCZ in boreal spring around the equator and in boreal summer around 9°N. In CM2.5, the spring location of the ITCZ is around 2°S and its summer location is around 9°N, bringing its meridional migration more in line with that observed relative to CM2.1. The excessive southward migration bias of the ITCZ in CM2.1 and CM2.5 during January–March is associated with the weak bias of southerly winds stress along the West African coast (Table 1b).
The excessive northward migration bias of the ITCZ in CM2.1 during boreal summer is linked with the seasonal variation of SST and the subsurface Guinea Dome region in the northeastern tropical Atlantic, a region where seasonal variations of SST are large. In June, the difference in the northeastern tropical Atlantic SST between CM2.1 and CM2.5 is very small and the maximum peaks of SST are located around 4°N in CM2.1 and CM2.5 (Fig. 5a). However, we found the large difference in August (Fig. 5b). CM2.1 shows colder SST relative to CM2.5 and observations within 5°–15°N. As a result, CM2.1 shows double peaks of SST and the maximum SST around 16°N. This SST bias leads to the excessive northward migration bias of the ITCZ in CM2.1 in boreal summer.
To understand the mechanisms behind the annual cycle of the northeastern tropical Atlantic SST, we build a diagnostic mixed-layer heat budget (appendix). The annual cycle of the SST in CM2.1 in boreal summer is significantly different from that in CM2.5 and observation. This difference can be largely understood in terms of the ocean dynamical contribution (Table 2). The cooling tendency from ocean dynamics in boreal summer is due to subsurface doming of the thermocline in the northeastern tropical Atlantic, known as the Guinea Dome (or the Dakar Dome, Mazeika 1967). Figure 6 shows the summer field of wind stress, Ekman upwelling, and ocean stratification around mixed layer depth. Ekman upwelling is calculated for wind stress in observations and models. In the observational estimates, the Guinea Dome is climatologically located within 10°–15°N, 35°–20°W. The ocean stratification over the simulated dome in CM2.1 is four times stronger than observation, and the simulated dome in CM2.1 is located farther north by 4°, and extends farther westward relative to observation. The strong bias in the Ekman upwelling over the Guinea Dome in boreal summer influences the seasonal variation of SST in this region. The mechanism is shown schematically in Fig. 7. The Guinea Dome develops from boreal spring through summer from the Ekman upwelling associated with the northward migration of the ITCZ (Siedler et al. 1992; Yamagata and Iizuka 1995; Doi et al. 2009), which cools the mixed layer through entrainment (Doi et al. 2010). This is consistent with the observational study of Yu et al. (2006). CM2.1 overestimates the cooling tendency of the SST by the ocean dynamics associated with the cold subsurface Guinea Dome and strong Ekman upwelling. CM2.5 significantly improved those biases and successfully captures the coupled process among the northward migration of the ITCZ, the SST, and the subsurface Guinea Dome.
Although the equatorial region and the Angola–Benguela area also show large seasonal variations in observation and two models, we focused on the northern tropical Atlantic in this paper because the significant improvement in CM2.5 is found in the northward migration of the Atlantic ITCZ. In contrast to the northern tropical Atlantic, the differences between CM2.1 and CM2.5 did not result in an improvement to the serious SST biases in the equatorial region and the Angola–Benguela area (Fig. 1).
4. Interannual variation
Maps of the standard deviation of the interannual SST anomaly are shown in Fig. 8. As in the seasonal variation of SST, there are three regions of large interannual variability, each of which has strong seasonal phase locking of the variability (Fig. 9). Thus, interannual variations in these regions of the tropical Atlantic can be described as a modulation of the annual cycle. In this paper we focus on the seasonal phase locking of the interannual variations of SST averaged in the northern tropical Atlantic (NTA) between 35° and 20°W, 5° and 20°N because CM2.5 shows a significantly reduced bias in the seasonal phase locking of interannual variations in the NTA relative to CM2.1 (Fig. 9a). However, in the equatorial Atlantic (ATL3) and the Angola–Benguela area (ABA) both CM2.1 and CM2.5 fail to simulate the seasonal phase locking of the interannual variations of SST.
The observed standard deviation of the NTA SST anomaly develops from February, reaches its maximum peak in April, and decays abruptly from May through September. The observed standard deviation of the SST anomaly in September is smaller by ~50% than that in April. CM2.1 successfully simulates the development phase of interannual anomalies from early winter through boreal spring. However, the standard deviation in CM2.1 keeps increasing March through May, and the decay from June through August is substantially weaker than that observed. As a result, the interannual variations of August NTA SST in CM2.1 are stronger by ~65% relative to observations. This bias in the seasonal phase locking of the interannual variation of NTA SST is significantly reduced in CM2.5. Since the interannual variability of NTA SST in boreal summer may impact Sahel rainfall, the West African monsoon, and Atlantic hurricane activity, the CM2.1 bias in the seasonal phase locking of the interannual variations of NTA SST could represent a serious limitation to seasonal predictions based on CM2.1. Therefore, we aim to understand its causes in CM2.1 and why this bias is improved in CM2.5.
We explore a composite analysis to help understand the mechanism of the seasonal phase locking of interannual variations of NTA SST. We construct a composite by averaging, based on selecting warm (cold) SST years in the NTA, when the interannual SST anomaly in the NTA exceeds one standard deviation during the mature season of March–May. The details are shown in Table 3. We have about 1.5 yr (decade)−1 as a composite year in observations and models.
In observations, positive SST anomalies develop in the NTA from early winter through April and the warming tendency is strongest in February (figure not shown), mainly due to the surface enthalpy flux contribution, , (Table 4). This feature is well simulated in CM2.1 and CM2.5. We note that interannual variations of surface enthalpy flux contribution, (see appendix), include not only interannual variations of surface enthalpy flux, but also interannual variations of mixed layer depth (e.g., Morioka et al. 2010, 2011). However, we have confirmed that interannual variations of mixed layer depth do not contribute to over the northern tropical Atlantic in boreal spring, and interannual variations of surface enthalpy flux Q are dominant (figure not shown). All observational datasets, CM2.1, and CM2.5 show the wind-induced latent heat flux anomaly as the dominant term in the net surface enthalpy flux anomaly (Table 4). The composites of the latent heat flux and wind stress anomalies in February (Fig. 10) suggest that the warming mechanism is consistent with the WES positive feedback (Xie 1999). The WES feedback is associated with the southwesterly wind anomaly and weaker trade winds in the northern tropics. Those wind anomalies result in less evaporation and, thus, suppressed latent heat loss from ocean, leading to warmer SST in the northern tropical Atlantic. The dominance of this mechanism in the growth of SST anomalies in the northern tropical Atlantic has been discussed in previous works (Carton et al. 1996; Chang et al. 1997; Xie 1999; Huang and Shukla 2005; Hu et al. 2008; Lin et al. 2008). We confirmed that CM2.1 and CM2.5 reasonably capture the WES feedback and there are no significant differences between CM2.1 and CM2.5 (Fig. 10), although some biases are found in the two models. In both models, the strongest warming area is about 4° southward from that in observations. This may be due to the southerly ITCZ location bias during boreal winter–spring in the mean climatology, as discussed in section 3.
Although the essence of the boreal spring development mechanism is reproduced in both CM2.1 and CM2.5, the boreal summer decay mechanism is incorrectly represented in CM2.1. In August of warm NTA years, the difference in the northeastern tropical Atlantic SST between CM2.1 and CM2.5 is much larger relative to February. The largest difference in SST anomalies between CM2.1 and CM2.5 is located between 8° and 12°N in August (Fig. 11)—where the Guinea Dome is located in early summer. In this region, the rate of change of the SST anomaly in CM2.1 is different from observation and CM2.5: CM2.1 shows that the warming tendency of SST is stronger from February through May. In particular, during April the positive SST anomaly in CM2.1 still develops strongly, while the SST anomalies in the observations and CM2.5 start to decay in April. We explored the diagnostic mixed layer heat budget anomaly in this region [Eq. (A2) in the appendix, Table 5]. Although there are substantial discrepancies among observational datasets, the NCEP–NCAR reanalysis data indicates that the ocean dynamical contribution plays an important role in the summer decay. Some previous works suggested that the ocean dynamical cooling counteracts warming by the WES feedback and provides an important negative feedback that helps to terminate warm events (Joyce et al. 2004; Lee and Wang 2008; Doi et al. 2009, 2010; Mahajan et al. 2010). However, CM2.1 fails to simulate the cooling effect by the ocean dynamical contribution and shows a warming tendency by the ocean dynamical contribution. Therefore, the peak season of the SST anomaly moves from boreal spring into early summer and the warm SST anomaly is sustained through late summer in CM2.1. The bias in the ocean dynamical contribution is significantly reduced in CM2.5
Why does CM2.1 fail to simulate the decay mechanism by the ocean dynamical contribution in boreal summer? In large part, the answer lies in the inability of CM2.1 to capture the negative feedback associated with the Guinea Dome. In the diagnostic mixed layer heat budget anomaly of Table 5, the vertical entrainment contribution induced by the entrainment rate anomaly and climatological ocean stratification,
(appendix), can explain about 80% of the CM2.1 bias of the warming tendency during boreal early summer. As discussed above, the warm SST anomalies in the NTA is associated with an anomalously northward migration of the ITCZ, which leads to a positive wind stress curl anomaly and strong Ekman upwelling anomaly over the climatological Guinea Dome region. Therefore, enhanced upwelling plays an important role in the termination of the warm SST anomaly in the NTA through entrainment (schematic diagram, Fig. 18 of Doi et al. 2010). Observations show the Ekman upwelling anomaly between 6° and 15°N over the climatological Guinea Dome region associated with the northward migration of the ITCZ (Figs. 12a and 13). However, in CM2.1, there is a downwelling Ekman anomaly between 3° and 15°N (Figs. 12b and 13). This disagreement as to the sign of the wind-induced vertical velocity in 3°–15°N between observations and CM2.1 is due to the climatological mean difference in the location of the ITCZ and the characteristic of Ekman upwelling around the ITCZ. In the mean climatology, CM2.1 shows the stronger Ekman upwelling and broader zone of Ekman upwelling within 6°–18°N around the simulated ITCZ relative to observation (Fig. 13).
The CM2.1 bias for an excessive seasonal northward migration of the ITCZ in CM2.1 from boreal spring to summer (discussed in section 3) is amplified in the interannual time scale, and results in the Ekman upwelling anomaly further north around 18°–25°N and the Ekman downwelling anomaly within 3°–15°N. The downwelling anomaly over the strong doming of thermocline results in persistent warm SST anomalies in CM2.1 between 8° and 12°N during August. This incorrect oceanic role associated with the subsurface ocean in CM2.1 is significantly improved in CM2.5 in part because the seasonal meridional migration of the ITCZ and the Ekman upwelling in the northern tropical Atlantic is well simulated (Figs. 12 and 13). Note that the centers of the simulated domes in warm NTA years in CM2.1 and CM2.5 are similar to the mean climatologies in CM2.1 and CM2.5 because anomalies of ocean stratification around mixed layer depth are less than 5% of the mean climatology. Recovering the correct seasonal meridional migration of the Atlantic ITCZ is key not only for the mean annual cycle, but also for the seasonal phase locking of the interannual variations of NTA SST.
The cold years of the NTA SST can be explained by using similar mechanisms of opposite sign to the warm years, although there are some differences in the effectiveness of the entrainment cooling and statistical significance. (Because of page limits, relevant figures are not shown in this paper.)
5. Discussion and summary
Using output from the “1990 Control” simulations of two coupled GCMs (CM2.1 and the high-resolution CM2.5), the tropical Atlantic biases in mean state, annual cycle, and interannual variations were investigated. Many aspects of the simulation are improved in CM2.5—yet biases persist. The mean annual cycle of rainfall over the Sahel and northern South America are well simulated in CM2.5, while CM2.1 shows excessive rainfall over the Sahel and deficient rainfall over northern South America, particularly in boreal summer. Improvements in simulation of rainfall in CM2.5 are tied to a more realistic meridional migration of the model’s ITCZ. In CM2.1, the meridional migration of the ITCZ is larger than observed.
Biases in the meridional migration of the ITCZ in CM2.1 arise from a coupled evolution of SST, Ekman upwelling velocity, and the subsurface Guinea Dome. The Guinea Dome develops from late boreal spring through summer and matures in boreal autumn, driven by the wind-induced Ekman upwelling associated with northward migration of the ITCZ. CM2.1 shows excessive Ekman upwelling and an enhanced Guinea Dome associated with farther northward migration of the ITCZ. Entrainment cooling over the Guinea Dome plays an important role in seasonal variation of the upper SST from boreal summer through autumn. CM2.5 produces an improved local wind-induced Ekman upwelling and oceanic stratification over the Guinea Dome in boreal summer. The coupled process connecting the ITCZ, SST, and the subsurface Guinea Dome strongly influences on the seasonal dependence of the interannual variations of SST in the northern tropical Atlantic.
The interannual SST anomaly in the northern tropical Atlantic develops from early boreal winter through spring, and reaches a maximum in April in observations. There are no gross differences in the simulation of the development mechanism between CM2.1 and CM2.5. However, the mechanisms for summer decay of the NTA interannual variability is not well simulated in CM2.1. CM2.1 shows persistent warm SST anomalies between 8° and 12°N through August in the warm NTA years, when observations show no significant SST anomalies. The bias in the summer decay phase of the interannual SST anomaly in CM2.1 is mainly due to the ocean dynamical contribution by the subsurface Guinea Dome. In observational estimates, warm SST anomalies over the subsurface Guinea Dome are reduced by a negative feedback tied to the Guinea Dome (Doi et al. 2010): an anomalously northward migration of the ITCZ associated with the warm SST anomaly in the northern tropical Atlantic leads to a cyclonic wind stress curl anomaly, and thus enhanced Ekman upwelling over the Guinea Dome region. This coupled evolution plays a critical role in the summer decay of the warm SST anomaly through entrainment cooling (schematic diagram is shown in Fig. 14a). This mechanism is also interpreted as an enhanced annual cycle shown in Fig. 7. However, CM2.1 fails to simulate this negative feedback. In CM2.1, the climatological ITCZ is located to the north of that observed and it shows excessive Ekman upwelling and a broad zone of Ekman upwelling within 6°–18°N around the simulated ITCZ. These characteristics lead to an Ekman upwelling anomaly between 18° and 25°N and the Ekman downwelling anomaly between 3° and 15°N during warm NTA periods. The Ekman downwelling anomaly over the strong doming of thermocline sustains the warm SST anomaly between 8° and 12°N through August (Fig. 14b).
Meanwhile, CM2.5 successfully reproduces the seasonal phase locking of the interannual variations of the northern tropical Atlantic SST. This is due to a more realistic climatological meridional migration of the ITCZ in CM2.5, which leads to a more realistic positioning and strength of the Guinea Dome. This improved mean state provides a background for a more realistic decay of interannual anomalies in the northern tropical Atlantic. The coupled processes that connect northward migrations of the ITCZ, SST, and the subsurface Guinea Dome are key to understand tropical Atlantic variability.
We hypothesize that CM2.5 may exhibit better prediction skill of northern Atlantic climate conditions and their impacts than CM2.1 because CM2.5 more successfully reproduces the annual mean and annual cycle of rainfall over the Sahel and northern South America, the subsurface Guinea Dome variations, and the seasonal phase locking of the interannual variations of the northern tropical Atlantic. Since the seasonal phase locking of anomalies in the NTA is key to its influence on seasonal phase-locked phenomena (e.g., tropical cyclone activity), we hypothesize that the experimental predictions of seasonal hurricane activity described in Vecchi et al. (2011), which have been based in part on predictions for the NTA using CM2.1, will be improved by using CM2.5.
Differences between CM2.1 and CM2.5 are not only horizontal resolution but also include some changes to parameterizations, numerics, and the land model. At this stage, we cannot confidently assess whether increased resolution or changes to numerics and land model have been dominant in reducing the biases in precipitation and the ITCZ in CM2.5. Sensitivity experiments for exploring this question are being conducted as part of our current research activities.
Meanwhile, the differences between CM2.5 and CM2.1 did not lead to reductions in very large SST biases in the eastern equatorial region and Angola–Benguela area, which are found in almost all CMIP3 models (Richter and Xie 2008). Both the annual cycle and interannual variations of SST in these two areas are stronger in both models than in observations. In addition, neither model can simulate the seasonal phase locking of the interannual variations of SST over the Angola–Benguela area. Therefore, neither model realistically simulates the Atlantic Niño and the Benguela Niño, which are two major climate modes in the tropical Atlantic. A tendency for the weak southerly winds along the southwestern African coast already appears in the atmospheric components of these coupled models when forced with observed SST (Table 1), suggesting that atmospheric biases are likely causative factors for the emergent coupled biases. The weak bias of southerly winds stress along the West African coast is associated with the excessive southward migration bias of the ITCZ in CM2.1 and CM2.5 during boreal winter–spring. At this stage, we speculate that the problems are mostly related to atmospheric physics associated with deep convection and cloud processes in the AGCM. Sensitivity experiments for reducing the tropical Atlantic biases are also being conducted as part of the GFDL research program.
In this paper, we have focused on the tropical Atlantic basin. However, uncertainty still remains as to remote effects of the Pacific, the midlatitude Atlantic, and the tropical southern Atlantic on climate conditions in the tropical Atlantic (see reviews by Xie and Carton 2004; Chang et al. 2006). In particular, Czaja (2004) suggested that the seasonal dependence of the interannual variability in the northern tropical Atlantic is a reflection not only of local air–sea coupling but also the remote forcing by the North Atlantic Oscillation and the ENSO. Although the amplitude of the Pacific ENSO is stronger in CM2.1 than that in observations (Wittenberg et al. 2006), this bias is partially reduced in CM2.5 (Delworth et al. 2012). Exploring the relation between the tropical Atlantic and other basins is also under way.
We thank Drs. Andrew Wittenberg, Stephen Griffies, Rong Zhang, Gabriel Lau, Rym Msadek, Ian Lloyd, Syukuro Manabe, and Bruce Wyman for helpful comments and suggestions. We are grateful to the GFDL CM2.1 and CM2.5 modeling services team for their assistance with model infrastructure support and data processing.
Mixed Layer Heat Budget Analysis
We explore the diagnostic bulk mixed layer heat budget:
Here Tmix is the mixed layer temperature, a proxy for SST, ρ is the typical seawater density (1025 kg m−3), Cp is the typical heat capacity of the seawater (3996 J kg−1 K−1), and Hmix the mixed layer depth, calculated as the depth at which the potential density becomes 0.125 kg m−3 greater than the surface density, as used by Levitus (1982). The quantity Q denotes the net surface enthalpy flux, and qsw is the downward solar insolation that penetrates through the bottom of the mixed layer. Thus, the first term on the right-hand side represents the influence of atmospheric thermal forcing. The ocean dynamical contribution is simply estimated by the difference between rate of change of the mixed layer temperature and the surface enthalpy flux contribution.
To understand the detailed ocean dynamical contribution in the Guinea Dome region, we explore the detailed mixed-layer heat budget:
The second term on the right-hand side represents the oceanic cooling associated with entrainment, where Went is the entrainment rate, and Te is the temperature of water entrained into the mixed layer and assumed to be the temperature 5 m below the mixed layer (e.g., Qu et al. 2001). The entrainment rate can be assumed by
where denotes the rate of change of the mixed layer depth, Wmb is the vertical velocity at the base of the mixed layer, and is the horizontal transport. If Went is negative, we assume Went = 0. This estimation of the oceanic entrainment cooling is a well-known diagnostic method (e.g., Hagos and Cook 2009). The third term represents the horizontal heat transport in the mixed layer.