Tropical AtlanticBiases in CCSM4

Semyon A. Grodsky,James A. Carton, and Sumant Nigam

April 27, 2011

To be submitted to the Journal of Climate

Department of Atmospheric and Oceanic Science

University of Maryland

College Park, MD20742

AbstractThis paper focuses on the tropical Atlantic biases in the control simulation of the Community Climate System Model version 4 (CCSM4). We find that local and remote biases in both, atmospheric and oceanic components of the coupled model contribute.Like in previous version, CCSM3, the atmospheric component of CCSM4 (CAM4) has abnormally high (by a few mbar) mean sea level pressure in the subtropical pressure highs and abnormally low in the polar lows. As a result all global scale winds are accelerated. Wind stress in the trade winds is approximately 0.05 Nm-2 (~2 ms-1) stronger.In spite of anomalously strong trade winds in the north and south, the SST bias in the tropical Atlantic changes sign across the Equator andprojects on the naturalmode of variability inherent to the tropical Atlantic.The strongest warm SST bias occurs in the Angola-Benguelafront. By comparing eddy resolving and eddy permitting ocean simulations with simulations on a CCSM4 ocean grid in the Benguela region, we find that decrease in horizontal resolution (below the local Rossby radius of deformation) impacts circulations in the eastern boundary. Besides the spatial resolution, a proper local wind forcing is equally important. If the low-level wind jet along the Benguela coast is not resolved by the atmospheric component, the coastal jet of the cold Benguela Current is replaced by a broad northward flow resulting in stretching of the SST front and warming of SSTs at latitudes where cold water transported by Benguela Current is normally present. This warm SST isexpanded seaward by ocean advection and is amplified in coupled simulations by positive feedback from marine stratocumulus. This study suggests that smaller biases in coupled simulations of the tropical Atlantic may be achieved via improving the large scale atmospheric pressure fields,better simulations of coastal winds,and by better simulations of the ocean boundary currents. Improvement of coastal winds and currents requires improvement in horizontal resolution beyond 1othat is currently operational in CCSM4.

1. Introduction

Although the climate of the tropical Atlantic Ocean is mostly seasonal, its coupled simulation still remains a problem that is evidentin notorious biases in regional winds and SST that have being present in recent generations of coupled models (Davey et al., 2002; Deser et al., 2006;Chang et al., 2007). In the Pacific, better representation of the deep-convection in the NCAR Community Atmosphere Model version 4 (CAM4) has lead to significant improvements in the phase, amplitude and spatial patterns of El Nino Southern Oscillation (Neale et al., 2008). But in the Atlantic improvements are not that noticeable. Due to proximity of continents the Atlantic coupled air-sea system is impacted by land processes (e.g. Zeng et al., 1996, Richter and Xie, 2008). Thus errors originated in any module of the Atlantic ocean-atmosphere-land system may impact each other and grow through coupled interactions.

Several recent studies have linked the causes for the persistent tropical Atlantic biases in coupled simulations with problems in the atmospheric component. In particular, Chang et al. (2007) have found that warm SST bias in the equatorial and southeastern tropical Atlantic in CCSM3 is due to below normal zonal equatorial winds during March-April-May (MAM) thatoriginate due to deficit of rainfall (lack of diabatic heating/ colder air/ higher pressure) in the Amazon. Thebias in equatorial westerlywinds during MAM is also present in uncoupled atmospheric model component, CAM3,forced by observed SST (Atmospheric Model Intercomparison Project, AMIP, style run) suggesting that the bias is initiated in the atmospheric component. The link between equatorial zonal winds and Amazon rainfall have been demonstrated by Chang et al. (2008) who have reproduced the equatorial westerly bias in a model forced by diabatic heating bias over the Amazon. In parallel, Richter and Xie (2008) have shown that erroneously weak Atlantic Walker circulation in conjunction with deficient and excessive terrestrial precipitation over equatorial South America and Africa, respectively, are robust across coupled modelsfrom the third Coupled Model Intercomparison Project (CMIP3) and their uncoupled AMIP counterparts.The causes of biases in tropical convection over the two continents may be in boundary conditions from the land component or in the convection scheme used by atmospheric component but further studies are needed to evaluate their relative importance (Richter et al., 2010b). Using a simplified model Zeng et al. (1996) have demonstrated that Atlantic Walker circulation weakens and equatorial zonal SST gradient drops by 1C as a result of changes in land albedo and evapotraspiration due to Amazon deforestation.

Westerly bias in equatorial winds during MAM leads to abnormaldeepening of the thermocline and warming of the cold tongue in the east inthe next season (JJA). In CCSM3 this abnormal thermocline deepening is accompanied by reversal of the Equatorial Undercurrent (EUC) and by reversal of normally westward gradient of equatorial SST (Chang et al., 2007). Thiserroneous, eastward SST gradient is further amplified and prolonged by the Bjerknes feedback and is present in all CMIP3 models (Richter and Xie, 2008) as well as in most of the non-flux-corrected coupled simulations analyzed by Davey et al. (2002). The important role of valid equatorial wind stress for proper simulations of the equatorial SST gradient has been confirmed by Wahl et al. (2009) andRichter et al. (2010b).In coupled simulations the cold tongue SST in JJA reduces by 3 K and warm pool SST increases by 0.5 K if model-generated wind stress in MMA is substituted byobserved climatology.

The warm SST bias extends from the equatorial zone into the tropical southeastern Atlantic where it is stronger, more persistent, and less seasonally dependent (Stockdale et al. 2006; Chang et al., 2007; Huang and Hu, 2007). In fact,the largest time mean SST biases develop along the eastern boundaries of subtropical gyres (Large and Danabasoglu, 2006). Many studies have linked this off-equatorial SST bias with biases in equatorial windsand their remote impact via the equatorial and coastal Kelvin waves (Florenchie et al. 2004, Richter et al. 2010a). But sensitivity experiments of Richter et al. (2010 a,b) have shown that errors in both zonal equatorial winds and local coastal winds (impact on local upwelling) contribute comparably to the warm SST biases in the southeastern tropical Atlantic.

But ocean model biases also contribute.In fact, the warm SST bias is present along the southwestern coast of Africa in uncoupled ocean simulations (Large and Danabasoglu, 2006).The dynamics of this southeastern Atlantic boundary region is strongly affected by the low–level atmospheric jet along the Benguela coast that is driven by the south Atlantic subtropical high and topographic enhancement of winds west of the Namibia highland (Nicholson, 2010). This coastal wind jet drives local upwelling and the coastal branch of the Benguela Current which northward flow is mainly the result of geostrophic adjustment to the upwelling (Colberg and Reason, 2006). The strength of the Angola-Benguela frontal zone isfound to be related to the strength of the southerly windstressthat controls the coastal upwelling and must be resolved by atmospheric forcing. Even with ‘perfect’ atmospheric forcing the horizontal grid spacing of ocean model component should resolve the local baroclinic Rossby radius of deformation (~ 40–60 km) to propersimulate the upwelling and resolve the cross-coastal sea levelgradient that maintains the equatorward coastal currents (Colberg and Reason, 2006).But, examining impacts of improved spatial resolution on coupled simulation of the eastern boundary doesn’t provide accorded conclusions. In particular, Tonniazzo et al. (2010) have found apparent improvements of SSTs in dynamically similar Peruvian upwelling region in high resolution run of the Hadley Center coupled model (HiGEM) with an eddy permitting 1/3o ocean and T144 (1.25o longitude x 5/6o latitude) atmosphere. But,Kirtman(2011, personnel communication)reports significant warm SST biasin the Benguela region in an eddy resolving 0.1 deg oceancoupled with a T288 CAM4. If the impact of anomalously weak Benguela upwelling(or any other reason behind the warm regional SST bias)is overridden by setting coastal temperature and salinity along the southeastern Atlantic boundary close to observations, the SST improves across the entire southeastern Atlanticin a coupled run with 1o ocean model that doesn’t explicitly resolve (Large and Danabasoglu, 2006).

Warm coastal SST bias is advected into open regions of the southeastern Atlanticby wind-driven ocean currents (Large and Danabasoglu, 2006). In addition, the marine stratocumulus clouds (developing over cold SSTs) provide yet another coupled ocean-atmosphere mechanism for extending warm SST anomalies off the coast.Marine stratocumulus clouds cover significant portion of the ocean and are particularly evident over the upwelling areas along the western coasts of continents (Zuidema et al., 2009). Due to their high reflectivity, stratocumulus clouds playimportant roles in the ocean radiation budget. Theyaffect SST not only through the radiative shading effect, but also dynamically: Long-wave cooling from the cloud tops is balanced by adiabatic warming, i.e., subsidence. The subsidence leads to near-surface divergence, and thus counter clockwise circulation in the Southern Hemisphere, i.e., to southerlies along the coast (see Nigam, 1997). Not enough clouds results in weaker upwelling favorable southerly wind along the coast and warmer SSTs. Thus, both, radiative and dynamic effectsof stratocumulus clouds provide positive feedback on SST. Anomalously warm SST reduces low-level cloud cover thus forcing SST warming in larger area and spreading the SST anomaly seaward in the direction of mean southeasterly winds (Huang and Hu, 2007).

Sea surface salinity (SSS) also affects SST and air-sea interactions through its impact on the upper ocean stratification and barrier layers (e.g. Breugem et al., 2008). Tropical SSS biases in coupled modelsare generally attributed to errors in precipitation. In CCSM3 the fresh SSS bias is the largest south of the equator (in excess of 1.5 psu) due to the southward shift of the ITCZ and the “double” ITCZ (Large and Danabasoglu, 2006).Meridional shift of rainfall has significant impact on the tropical rivers discharge in the Atlantic sector. In particular, the Congo dischargein CCSM3 more than doubles the climatological discharge of Large and Yeager (2009). An excessive river plume leads to overrepresentation of barrier layers (BLs) and relatedSST warming (e.g. Breugem et al., 2008) providing a positive feedback on already warm SST bias in the region. Similarly in the north tropical Atlantic, the underrepresentation of BLs (that are normally present in this area due to advection of the transformed Amazon freshwater overlying the subducted high salinity water from the subtropical SSS maximum) results in anomalous deepening and entrainment cooling of winter mixed layers that again provide a positive feedback on already cold SST bias in the region (Balaguru et al., 2010).

2. Model and Data

This research focuses on the tropical Atlantic biases in the Community Climate System Model version 4 (CCSM4).The CCSM4 is a coupled climate model composed of four separate models simultaneously simulating the earth's atmosphere, ocean, land surface and sea-ice, and one central coupler component[1].The CCSM4 data used in this study are the output data of the 1300 yr control model integration (archived as b40.1850.track1.1deg.006). This run is forced by historical ozone, solar, volcanic, green house gases, carbon, and sulfur dioxide/trioxide. Our analysis focuses on data for 97 year period (model years 863-959). A sensitivity examination has been carried out to ensure that the climatology of this particular period is similar to that of other periods. Our focus is on the performance of atmospheric and oceanic model components.

The atmosphere component of CCSM4, Community Atmosphere Model, version 4 (CAM4); employs an improved deep convection scheme (in comparison with CAM3 of Collins et al. 2006) by inclusion of convective momentum transport and a dilution approximation for the calculation of convective available potential energy (Neale et al., 2008). It is run on a 26 vertical levels, 1.25° longitude x 1° latitude grid. To separate errors originated in the atmospheric component from those in the coupled system we also use a stand alone CAM4 run on the same grid but forced by observed SST (CAM4/AMIP, f40.1979_amip.track1.1deg.001).

The ocean model component of CCSM4 is the Parallel Ocean Program version 2 (POP2) (Smith et al., 2010). Among otherimprovements the POP2implements a simplified version of the near-boundary eddy flux parameterization of Ferrari et al., (2008), vertically-varying thickness and isopycnal diffusivity coefficients (Danabasoglu and Marshall, 2007),modified anisotropic horizontal viscosity coefficients with much lower magnitudes than in CCSM3 (Jochum et al., 2008); and modified K-Profile Parameterization that uses horizontally-varying background vertical diffusivity and viscosity coefficients (Jochum, 2009). The number of vertical levels has been increased from 40 levels in CCSM3 to 60 levels in CCSM4. The ocean component of CCSM4 is run on a displaced pole grid with average horizontal resolution of 1.125°-longitude x 0.55°-latitude in midlatitudes. To separate errors originated in the ocean component from those in the coupled system we also use a stand alone POP/NYF run on the same grid and forced by repeating Normal Year Forcing (NYF) fluxes of Large and Yeager (2009), (c40.t62x1.verif.01).To explore impacts of grid resolution of the ocean model, the CCSM4 run is compared to an eddy permitting POP2 run (referred to as POP_0.25, Carton and Giese, 2008) forcedby atmospheric fluxes from the 20-th Century Reanalysis Project (CR20) version 2 of Compo et al. (2011). We also use an eddy resolving POP2 run (POP_0.1/NYF) of Maltrud et al. (2010) forced by repeating NYF.

For each variable the model biases are evaluated for each calendar month as the difference between model and observed climatology. A number of observation data are used for comparisons with the CCSM4 simulations. In choosing observations we require (if possible) a minimum 10 year records in order to estimate a stable seasonal cycle.Our SST data is the Reynolds et al. (2002) optimal interpolation version 2 analysis spanning late 1981 - onward. Observed 10m neutral winds are available from the QuikSCAT scatterometer (e.g.Liu, 2002) during mid-1999-2009. QuikSCAT wind stress is provided by Bentamy et al. (2008) and by climatology of Risien and Chelton(2008). For shortwave radiation (SWR) we rely on retrievals from the Moderate Resolution Imaging Spectro-radiometer (Pinker et al., 2009), which is available on a 1°x1° grid since mid-2002. Latent heat flux is based on the recent update of Institut Francais pour la Recherche et l’Exploitation de la Mer (IFREMER) weekly satellite-based turbulent fluxes of Bentamy et al. (2003, 2008) spanning 1992-2008. Precipitation is provided by the Climate Prediction Center Merged Analysis of Precipitation (CMAP) of Xie and Arkin (1997), which covers the period 1979 -present. Mean sea level pressure (MSLP) is the EuropeanCenter for Medium Range Weather Forecasts ERA-40 reanalysis (Uppala et al., 2005) monthly means spanning the period from 1958 through 2001. In-situ measurements from the PIRATA moorings in the tropical Atlantic (Bourles et al.,2008) are also used for comparisons.Ocean salinity is provided by the Simple Ocean Data Assimilation (SODA) of Carton and Giese (2008).

3. Results

The presentation of the results is organized in the following way. In the first part of this section we address errors in the large scale atmospheric circulation and compare them with errors in the tropical-subtropical Atlantic SST. We will see that wind errors have mostly symmetric pattern (trade winds are accelerated north and south of the Equator) while the SST errors resemble a pattern of the Atlantic meridional mode (cold north and warm south).This dipole-like SST error pattern suggests in turn that errors in the coupled model climate may be further amplified by projecting on the natural mode of variability inherent to the Atlantic (Chang et al., 2007). We next examine the reasons for the dipole-like pattern of SST errors and its link with deficiencies in the atmospheric and oceanic components of the coupled model.

Mean sea level pressure and surface winds. Over the off-equatorial latitudes, the excessive subtropical high pressure systems circle the globe (Fig. 1). The anomalous pressure pattern originates from imperfections in the atmospheric module that is evident by apparent similarity of the time mean MSLP bias in CCSM4 and its atmospheric module forced by observed SST(CAM4/AMIP, compare Figs. 1a, 1b). Biases in the global scale pressure fields are inherited by CAM4 from previous versions of the atmospheric module.In particular, similar patterns of anomalously strong subtropical high pressure systems are present in CCSM3 and its atmospheric module (CAM3/AMIP) (Figs. 1c, 1d). The CCSM4 subtropical high systems have larger bias in the Atlantic sector where MSLP high exceeds the normal by 4 to 5 mbar in the north and south. In spite of these biases, the MSLP has been improved in comparison with CCSM3. This improvement is particularly evident in the North Atlantic where the region of above normal MSLP has smaller spatial extension and is lower in magnitude (comp. Figs. 1a, 1c).Improvements in MSLP are noticeable in the Northern Hemisphere while they are less evident in the Southern Hemisphere. Moreover, the MSLP bias in the South Atlanticin CCSM4 and CAM4/AMIP is stronger than in corresponding version 3 runs. Anomalously strong meridional pressure gradientsare reflected in above normal wind stress (Fig. 2). In fact, all global scale wind systems are anomalously strong including north and south mid-latitude westerly winds and northeasterly and southeasterly trade winds. Above normal winds are expected to produce stronger evaporation and mixing, thus colder SST.

SST In spite of anomalously strong trade winds in the north and south, the SST bias in the Atlantic sector is asymmetrical with respect to the Equator (Fig. 3). In the northern tropics the time-mean SST is too cool by 1-2°C. This cooling is consistent with anomalously strong northeasterly trades, wind stirring,and evaporative heat loss. Similar cooling is present in the south. But, in distinction from the north where the cold SST biasstretcheszonally across the entire basin, the cold SST bias in the south is limited to the west, though stronger than normal winds are basin-wide there (Fig. 2). Despite stronger winds, the SST has a warm bias in the southeastern tropical Atlantic. As discussed in the Introduction, a number of local and remote wind biases affect SST in the southeastern tropical Atlantic. SST warming in the eastern boundary of the southeastern tropical Atlantic may be initiated by below normal southerly coastalwinds (that maintain local upwelling) orby remote impacts from the equatorial Atlantic. But the warm coastal SST biasis also present in stand-alone ocean run, POP/NYF, forced by observedair-sea flux climatology (Fig. 3). This suggests that ocean model component also contributes to SST bias in the southeastern tropical Atlantic. The meridional dipole-like SST bias in CCSM4 is persistent in all seasons (Fig. 4). Its seasonal variations are more noticeable in the south where the warm SST bias in CCSM4 peaks in JJA and minimizes in DJF in general correspondence with the seasonal cycle of the warm bias in POP/NYF.