Regional and zonal mean atmospheric energy budget
We first compare observed and CMIP5 multi-model climatological mean regional distributions of the radiative and combined latent and sensible heat flux terms in the atmospheric energy budget (Fig. 2a–i). The multi-model mean is determined by averaging fluxes from CMIP5 models 1–30 in Table 1 after interpolating each model’s output to a common grid of 1° latitude by 1° longitude, which is the spatial resolution of the CERES EBAF data. The observations fall within the model interquartile range in 26, 35 and 38 % of the regions for RA, QA, and \(\nabla \cdot F_{A}\), respectively. For in RA, these typically occur for regions having an absolute model-observed difference <5W m−2 (80 % of the cases), whereas for QA and \(\nabla \cdot F_{A}\), a similar percentage is realized for regions with an absolute difference <10 W m−2.
Regional patterns in \(R_{A}\) are similar between observations and models (Fig. 2a, b), although marked differences are apparent in marine stratocumulus regions off the west coasts of North and South America where the models underestimate atmospheric radiative cooling (Fig. 2c). The likely reasons include an underestimation of non-overlapped low cloud in CMIP5 models (Nam et al. 2012) and a dry bias in the marine boundary layer (John and Soden 2007). The models also show less cooling in middle and high latitudes in both hemispheres. Conversely, CMIP5 atmospheric radiative cooling is more pronounced in convective regions such as central Africa, the Pacific Warm Pool, ITCZ, and the Amazon. Consistent with these results, Li et al. (2013) found that CMIP5 models underestimate TOA reflected solar and overestimate outgoing longwave radiation in convectively active regions of the tropics due to an underestimation in the amount of total ice and liquid atmospheric water content.
Both the observations and models show larger combined latent and sensible heat fluxes over the subtropical oceans compared to other latitudes (Fig. 2d, e), with a maximum at about 20°S in the southern Indian Ocean. Intense heat flux gain also occurs in the warm western boundary currents, such as the Gulf Stream off the east coast of the United States, the Kuroshio Current near Japan, the Agulhas Current off the coast of South Africa, the Brazilian Current off of South America, and the East Australian Current. There is also qualitative agreement in regions with cooler sea surface temperatures, such as the northern Pacific Ocean and Southern Ocean, as well as over vast land masses in the northern hemisphere (e.g., Saharan desert, Asia, North America). The largest differences occur near the equator over the central and eastern Pacific Ocean as well as over the Indian Ocean (Fig. 2f). Differences greater in magnitude than 20 W m−2 also occur in the North Pacific Ocean between 30°N and 50°N and over the Indian Ocean between 20°S and 40°S. Regional patterns in \(\nabla \cdot F_{A}\) (Fig. 2g, h) show a divergence of heat transport equatorward of 40° latitude in regions influenced by deeper convective cloud and cirrus anvils such as the Indian ocean, western Pacific, the tropical Atlantic and equatorial Africa and South America. Convergence of energy in the atmosphere is evident at latitudes greater than 40°, where atmospheric radiative cooling generally dominates over atmospheric heating by latent and sensible heat fluxes. Local maxima in \(\nabla \cdot F_{A}\) are associated with the western boundary currents, while a minimum occurs in the eastern Pacific Ocean cold tongue region.
The atmospheric energy budget terms show remarkable hemispheric symmetry (Fig. 3a–i). Zonal mean hemispheric differences (vertical bars in Fig. 3) are generally less than 20 % of the zonal mean in the tropics and midlatitudes, and closely track one another poleward of 65°. There is a slightly greater contrast in \(\nabla \cdot F_{A}\) between the tropics and midlatitudes in the SH (Fig. 3g), implying a stronger tropical-to-midlatitude heat transport in the SH atmosphere compared to the NH. The stronger convergence of heat in the SH midlatitudes (Fig. 3g) is associated with both increased atmospheric radiative cooling (Fig. 3a) and less latent and sensible heating (Fig. 3d) compared to the NH. Frierson et al. (2013) argue that in response to the weaker poleward heat transport in the NH lower latitudes, the tropical mean atmospheric circulation transports energy from the NH to the SH via a southward cross-equatorial flow. At the same time, hemispheric symmetry in TOA radiation at midlatitudes (Fig. 2a in Frierson et al. 2013) implies a greater ocean heat transport into the NH midlatitudes. Overall, the CMIP5 multi-model mean results show a similar latitudinal dependence in each hemisphere as the observations (Fig. 3b, e, h). The largest discrepancy occurs in the deep tropics around 5°N, where differences are greater in magnitude than 10 W m−2 for Q
A
(Fig. 3f) and 5 W m−2 for \(\nabla \cdot F_{A}\) (Fig. 3i). These differences are primarily due to excessive Q
A
in the ITCZ region of the eastern Pacific Ocean region (Fig. 2f).
Energy budget hemispheric asymmetry and cross-equatorial heat transport
At the TOA, the remarkable hemispheric symmetry in absorbed shortwave (SW) radiation in the observations (Voigt et al. 2013; Stephens et al. 2015) is not replicated in the CMIP5 multi-model mean. While observations show a near-zero SW TOA contribution to HT
EQ
(Table 2), the multi-model mean indicates 1.7 W m−2 more absorbed SW radiation in the SH than in the NH, corresponding to 0.22 PW SH to NH HT
EQ
(Table 3). The longwave (LW) hemispheric asymmetry in observations and models is in better agreement, contributing approximately 0.2 PW to HT
EQ
. As a result, HT
EQ
from the CMIP5 multi-model mean is more than double the observed value. We note that there is significant variability amongst the individual CMIP5 models (Fig. 4a) (standard deviation of 0.33 PW). This large spread is mainly due to differences in SW TOA flux hemispheric asymmetry amongst the models (Fig. 5).
Table 2 Observation based radiative and non-radiative fluxes for global, SH, and NH, and the corresponding implied cross-equatorial heat transport
Table 3 CMIP5 multi-model mean radiative and non-radiative fluxes for global, SH, and NH, and the corresponding implied cross-equatorial heat transport
Except for LW, observed and CMIP5 multi-model mean within-atmosphere energy budget terms are consistent to 5 W m−2 when averaged globally and over each hemisphere (Tables 2, 3). Both the observations and CMIP5 multi-model mean results show stronger net atmospheric radiative cooling (R
A
) and combined latent and sensible heating (Q
A
) in the SH compared to the NH, primarily determined by lower latitudes (Fig. 3). Consistent with the notion that AHT
EQ
transports heat from the warmer to colder hemisphere, the hemispheric difference in R
A
implies an AHT
EQ
of 0.75 PW from the NH to the SH (Table 2). The SW component of R
A
hemispheric difference contributes 0.46 PW and the LW contributes 0.29 PW. The hemispheric difference in Q
A
implies an AHT
EQ
of 0.51 PW in the opposite direction, resulting in a 0.24 PW NH to SH AHT
EQ
. In contrast, hemispheric asymmetries in CMIP5 multi-model mean radiative and combined latent and sensible heating are nearly identical but with an opposite sign (Table 3), implying no AHT
EQ
. However, the individual CMIP5 models show a large spread in AHT
EQ
(Fig. 4b), with a standard deviation of 0.2 PW. In over half the CMIP5 models considered, the combined latent and sensible heat flux contribution to AHTEQ dominates over the radiative contribution (Fig. 6a), implying a net SH to NH AHTEQ. Of the models showing NH to SH AHTEQ, only 3 models (MPI models) fall within the observational uncertainty for both Q
A
and R
A
contributions.
Cloud, surface and aerosol properties play an important role in determining the SW and LW contributions to hemispheric radiative heating/cooling in the atmosphere. While LW radiative cooling is greater in the NH for clear-sky conditions, the opposite is true for all-sky (Fig. 5a). The sign reversal is associated with a larger cloud fraction in the SH (Stephens et al. 2015), a greater fraction of low clouds in SH, and more high clouds in NH (Fig. 5b). These results are based upon a four-year average of merged CALIPSO, Cloudsat, CERES and MODIS (CCCM) EditionB1 data (Kato et al. 2010). Since low clouds enhance LW radiative cooling of the atmospheric column (Kato 2009), a greater fraction of low clouds in the SH enhances LW radiative cooling relative to the NH. In the SW, radiative heating in the atmosphere is greater in the NH for both clear and all-sky conditions. This is due to a higher surface albedo in the NH with its greater land fraction, which enables more light reflected from the surface to be absorbed by the atmosphere. Furthermore, since pollution is greater in the NH, there are likely more absorbing aerosols in the NH to further increase atmospheric SW absorption. Another contributing factor is precipitable water, which is slightly greater in the NH according to ERA-Interim reanalysis.
At the surface, the CERES data suggest a global mean R
S
of 109 W m−2, compared to 106 W m−2 for the CMIP5 multi-model mean (Tables 2, 3). Since both are in near surface energy balance, values for Q
S
are similar. Stephens et al. (2012) estimated an even higher value of Q
S
(112 W m−2). Remarkably, when Q
S
is directly obtained from satellite retrievals and/or reanalysis, its value is 14–17 W m−2 lower than what is required to balance the radiative contributions (Stephens et al. 2012; Wild et al. 2013; Loeb et al. 2014). There is still much debate about whether our inability to close the surface energy budget in observations is due to an underestimation in precipitation/evaporation and/or an overestimation in net surface radiation (R
S
) (Stephens et al. 2012; Loeb et al. 2014; Behrangi et al. 2014). However, recent studies have shown that satellite-derived downward SW and LW radiative fluxes are consistent with ground observations to within a few W m−2 over both land and ocean (Kato et al. 2013; Loeb et al. 2014; Rutan et al. 2015). It is thus unlikely that a large positive bias in RS is the reason for our inability to close the surface energy budget in observations.
The SH and NH hemispheric mean surface energy budgets show how the surplus of energy associated with the planetary imbalance of 0.6 W m−2 is distributed (Table 2). On average, the SH surface receives an extra 2.3 W m−2 and the NH surface loses 1.1 W m−2. Assuming the hemispheric asymmetry in ocean heat storage is much smaller (e.g. Durack et al. 2014; Drijfhout et al. 2014), this implies a 0.44 PW SH to NH OHTEQ (Frierson et al. 2013; Marshall et al. 2013). The hemispheric asymmetry in R
S
implies an OHTEQ of 0.95 PW from the SH to the NH and this is counteracted by the hemispheric asymmetry in Q
S
, which implies an OHTEQ of 0.51 PW from the NH to the SH. The net downward SW and LW radiative fluxes contribute equally to the hemispheric contrast in R
S
owing to a larger surface albedo and higher surface temperatures in the NH, which cool the surface more effectively. Since downwelling SW at the surface is equal in both hemispheres, the net surface SW contrast between the hemispheres is entirely due to surface albedo. With the exception of the IPSL models and MIROC4 h (models 19–22), which show zero OHT
EQ
, all CMIP5 models show a SH to NH OHTEQ, but only 8 models fall within observational uncertainty (Fig. 4c). In 20 of the models, SH to NH radiative and NH to SH latent and sensible heat contributions are overestimated compared to the observations by 0.1 PW or more. Consequently, it is quite feasible for a model to have large biases in both the radiative and combined latent and sensible heat components yet still provide the correct OHTEQ. This fortuitous cancelation of error masks more serious problems, some of which we explore further in the next section.
Revisiting the tropical precipitation asymmetry problem in climate models
Many studies have investigated aspects of the relationship between AHT
EQ
and the position of the ITCZ and/or differences in tropical precipitation between the NH and SH (Yoshimori and Broccoli 2008; Frierson et al. 2013; Hwang et al. 2013; Donohoe et al. 2013; Frierson et al. 2013; Marshall et al. 2013). Contrary to observations, which show a mean ITCZ at 6°N, more precipitation in the NH than in the SH, and a southward AHTEQ, climate models exhibiting a double-ITCZ and excessive SH tropical precipitation display a northward AHTEQ. Idealized model experiments indicate that when perturbations in one hemisphere are imposed through thermal forcing or changes in surface albedo, the ITCZ and tropical precipitation maximum shift towards the warmer/darker hemisphere (Chiang et al. 2003; Broccoli et al. 2006; Kang et al. 2008, 2009; Voigt et al. 2014). Because the Hadley circulation in the deep tropics governs atmospheric energy transport, a displacement of the circulation towards the warmer hemisphere is required in order to transport heat away from the warmer hemisphere across the equator via the upper branch of the Hadley circulation (Yoshimori and Broccoli 2008; Frierson and Hwang 2012).
To further explore these ideas in the context of the framework used in the previous sections, we examine the hemispheric asymmetry in RS and QA as a function of the tropical precipitation asymmetry index (TPA index, hereafter), defined as the NH minus SH precipitation difference divided by the tropical mean precipitation for latitudes equatorward of 20° (Hwang and Frierson 2013). The majority of models with excessive tropical precipitation in the SH (negative TPA index) overestimate net downward radiation at the surface and flux too much latent and sensible heat from the surface to the atmosphere in the SH relative to the NH (Fig. 7a, b). In addition, many of the models also underestimate the SH–NH contrast in atmospheric radiative cooling (Fig. 7c). This in turn leads to excessive heating of the atmosphere and cooling of the surface in the SH. Indeed, Hwang and Frierson (2013) find that in CMIP5 models with a negative TPA index, the tropical mean surface air temperature in the NH is similar to or less than that in the SH, whereas the opposite is true in observations and CMIP5 models with a positive TPA index. Associated with the excess heating of the SH atmosphere is a northward AHTEQ (Fig. 7d). In order to achieve the anomalous northward AHTEQ, the Hadley Circulation and ITCZ are displaced southward. For the CMIP5 multi-model mean, the excessive SH–NH contrast in combined latent and sensible heating contributes 60 % to the overall bias in atmospheric cross-equatorial heat transport, while the underestimation in SH–NH contrast in atmospheric radiative cooling contributes 40 %.
Relationship to SW and LW surface flux biases
Both the SW and LW components contribute to hemispheric asymmetry biases in RS in CMIP5 models that overestimate precipitation in the SH. Figure 8 shows the hemispheric asymmetry (SH minus NH) in RS and the corresponding SW and LW contributions for models with negative and positive TPA index. CMIP5 models with negative TPA index overestimate the hemispheric asymmetry in RS compared with CERES by 1.9 W m−2, with the SW component contributing 1.2 W m−2 (63 %) and the LW component contributing 0.7 W m−2 (37 %). In contrast, the average hemispheric asymmetry in RS for CMIP5 models with positive TPA index is within 0.1 W m−2 of CERES, and both the SW and LW components fall within 0.31 W m−2 of the corresponding CERES values.
The positive hemispheric asymmetry bias in RS for CMIP models with negative TPA index occurs from 10° to 70° (Fig. 9b). There is a positive bias in downward SW radiation in the SH subtropics and extratropics due to an underestimation in SW reflection by clouds (Figs. 9e, 11a). Consistent with these results, Hwang and Frierson (2013) show that models with a double ITCZ have too weak a SW cloud radiative effect over the Southern Ocean due to an underestimation in cloud fraction and/or cloud optical thickness. However, results in Fig. 10e show that the cloud bias also occurs at latitudes as far north as 10°S. Regionally, a positive model bias occurs over each ocean basin in the SH, and is especially large in marine stratocumulus regions (Fig. 11a). Interestingly, CMIP5 models with positive TPA index show a negative bias in zonal mean downward SW radiation in the SH compared to CERES equatorward of 50°S (Figs. 9f, 10f), implying too much cloud reflection. This mainly occurs over the Pacific Ocean between the International Date Line and 90°W, and over the Atlantic Ocean (Fig. 11b), regions associated with trade cumulus. In the NH subtropics and midlatitudes, SW model regional biases are generally positive over land and negative over ocean, except over marine stratocumulus off the coast of North America. Thus, negative regional biases over much of the north Pacific Ocean for CMIP5 models with a positive TPA index (Fig. 11b) are largely offset by positive biases over land at the same latitudes, resulting in relatively small NH zonal mean biases (Fig. 9f).
Equatorward of 10° latitude, the CMIP5 models overestimate net downward SW radiation at the surface in the NH (Fig. 9e, f), which contributes to a negative hemispheric asymmetry bias in RS (Fig. 9b, c). Regionally, this is due to cloud biases associated with the ITCZ over the central and Eastern Pacific Ocean. Regional biases are also quite large equatorward of 10°S, but substantial cancellation of error across longitudes reduces the zonal mean bias. Poleward of 60°S, both sets of CMIP5 models overestimate net SW downward radiation (Trenberth and Fasullo 2010; Bodas-Salcedo et al. 2014).
The LW contribution to the bias in hemispheric asymmetry is mainly confined to latitudes equatorward of 30° for CMIP5 models with negative TPA index (Fig. 9h). In both hemispheres the surface LW radiative cooling is overestimated compared to CERES, but the bias is stronger for 0°–30°N, resulting in a positive bias in hemispheric asymmetry. Regionally, the model biases between 0° and 30°N are dominated by the Sahel and India (Fig. 11c) with relatively small biases over the tropical oceans where sub-cloud LW emission by water vapor dominates. Results for CMIP5 models with positive TPA index show a marked overall improvement, particularly over the ocean away from the equator (Fig. 11d). Biases in downward LW radiation are still negative, but their magnitude is smaller and exhibits less hemispheric asymmetry (Fig. 9i).
Thus, in CMIP5 models with negative TPA index the model bias in RS hemispheric asymmetry is primarily due to excessive SW surface radiation poleward of 30°S (Fig. 9e) and too much LW surface radiative cooling in the NH tropics equatorward of 30°N compared to the same latitudes in the SH (Fig. 9h). In contrast, SW and LW biases in hemispheric asymmetry for CMIP5 models with positive TPA index largely cancel (Fig. 9f, i), resulting in a much smaller net bias in RS hemispheric asymmetry (Fig. 9c).