Volume 11, Issue 7 p. 2163-2176
Research Article
Open Access

Connecting Direct Effects of CO2 Radiative Forcing to Ocean Heat Uptake and Circulation

Molly E. Menzel

Corresponding Author

Molly E. Menzel

Department of Atmospheric and Oceanic Sciences, McGill University, Montreal, Quebec, Canada

Department of Earth and Planetary Sciences, Johns Hopkins University, Baltimore, MD, USA

Correspondence to: M. E. Menzel,

[email protected]

Search for more papers by this author
Timothy M. Merlis

Timothy M. Merlis

Department of Atmospheric and Oceanic Sciences, McGill University, Montreal, Quebec, Canada

Search for more papers by this author
First published: 19 June 2019
Citations: 5

Abstract

The ocean's response to direct atmospheric effects of increased carbon dioxide's (CO2) radiative forcing is examined. These direct effects are defined as the climate changes that result from forcing on a fast time scale of about a year, independent of the slower surface warming that the forcing also provokes. To evaluate how these direct effects impact ocean heat uptake and circulation, output of atmospheric general circulation model (GCM) simulations are used to force an ocean GCM with comprehensive boundary conditions. Perturbation simulations with the prescribed response to a quadrupling of atmospheric CO2 include altered surface winds, freshwater fluxes, downwelling shortwave radiation, and downwelling longwave cloud radiative effect. The perturbation simulations show that the intensification and poleward shift of surface winds, particularly in the Southern Ocean, strengthen the shallow overturning circulation in the tropical Pacific and deep overturning in the Atlantic. This, in turn, has a cooling effect on the global ocean at shallow depths. A two-layer energy balance model, designed to capture transient global mean climate change, is adapted to account for the altered ocean heat uptake from direct effects. The direct change in global mean ocean heat uptake is a decrease of about 0.3 W/m2 for quadrupling of CO2, offsetting about 5% of the surface longwave forcing.

Key Points

  • The direct effects of CO2 on the surface atmospheric fields that interact with the ocean are analyzed
  • The influence of direct effects on ocean heat content and circulation in ocean general circulation model (GCM) simulations is assessed
  • GCM results can be emulated by adapting two-layer energy balance model of ocean heat uptake to include direct effects

1 Introduction

The global climate's transient response to radiative forcing is strongly tied to the ocean's ability to uptake and store heat. As surface temperatures rise due to increasing greenhouse gases, the ocean absorbs significant amounts of heat, which mitigates the global mean warming at the surface. Therefore, uncertainty in the underlying processes governing the ocean response to radiative forcing is an important scientific challenge.

Much of the research designed to quantify the oceanic response to climate change makes use of coupled, atmosphere-ocean general circulation models (GCMs). These models are used to make quantitative projections of future climate and typically project an ocean heat uptake (OHU) that is about 90% of the Earth system's total change in energy content (Stocker et al., 2013). This OHU can be decomposed into a “passive” response to the altered surface boundary conditions, where the climatological ocean circulation advects the perturbed temperature field, and an “active” response, where the change in ocean circulation due to forcing redistributes the ocean's existing heat content (Garuba et al., 2018; Marshall et al., 2015; Winton, Adcroft, et al., 2013). The active component of OHU is, in part, related to a projected slowing of the Atlantic Meridional Overturning Circulation (AMOC), a robust projection under radiative forcing (Cheng et al., 2013) that is, in turn, related to changing ocean density gradients (e.g., Butler et al., 2016; Jansen et al., 2018). Here, we explore a missing pathway of ocean changes: the connection between direct atmospheric effects of radiative forcing and OHU. Traditionally, the climate's response to increasing greenhouse gases has largely been described relative to surface warming. However, recent work has revealed that there are “direct effects” (other names include “adjustments,” or fast responses) to radiative forcing that are independent of a change in temperature (Bony et al., 2013; Deser & Phillips, 2009; Grise & Polvani, 2014; Merlis, 2015; Sherwood et al., 2015). These direct effects arise from changes in the atmosphere's radiative fluxes that happen in proportion to the changes in the forcing agent, rather than in proportion to surface warming. For instance, increased CO2 concentration impacts the general circulation of the atmosphere, which in turn alters both the hydrological cycle and strength of the winds. In addition, direct responses to changes in CO2 can alter clouds, which affects the radiation budget of the atmosphere and surface (Andrews, Gregory, Forster, et al., 2012; Colman & McAvaney, 2011; Gregory & Webb, 2008; Huang, 2013).

How do these direct responses of the atmosphere to increased CO2 affect the ocean? Although there is a growing body of research investigating the direct response to increased levels of CO2 in the atmosphere, little work has been done to investigate how these direct atmospheric effects impact the ocean. However, it has been noticed that the ocean responds quickly to idealized forcing in years before a substantial amount of warming has been realized. In evaluating the ocean response to abrupt quadrupling of CO2 in coupled GCM simulations, Rugenstein et al. (2016) examined changes that occur within the first few years and how those patterns differ from the long-term responses. They found that an anomalous divergence at the equator causes a nearly instantaneous, albeit brief, cooling in that region and a warming at other latitudes. Rugenstein et al. (2016) label this short-time scale ocean behavior as an “adjustment” to the forcing, as it causes further changes to the atmospheric radiative forcing. They also found a short-term (1–2 years) strengthening of the AMOC before a long-term decline. Additionally, previous work has assessed the response of ocean heat content to surface flux changes that are projected to occur under global warming (e.g., Gregory et al., 2016; Saenko et al., 2015), although these studies have primarily used forcing scenarios such that the perturbation fields include both the direct and temperature-dependent response. While these simulation results reveal interesting behavior, there has not been a focused examination of how the direct atmospheric responses to radiative forcing impact ocean circulation and heat uptake. Here, we take a hierarchical approach to isolate and model these ocean responses.

To do so, we start by defining and analyzing what these direct effects are in the atmosphere (section 2). Then, we use them to perturb an ocean GCM with comprehensive Earth boundary conditions (section 3) to quantify their influence on OHU (section 4). Ocean GCM simulations have proven useful to understand aspects of transient climate changes and can capture the broad-scale behavior of coupled atmosphere-ocean GCM simulations of climate change (Armour et al., 2016; Huber & Zanna, 2017; Jansen et al., 2018). Motivated by the ocean GCM output, we propose adaptations to the two-layer Energy Balance Model (EBM; section 5) that incorporates the impact of the atmospheric direct effects on OHU.

2 Direct Effects of CO2 at the Air-Sea Interface in CMIP5

We define the direct effects of radiative forcing to be the immediate changes in atmospheric fields independent of the surface temperature changes (Bony et al., 2013; Grise & Polvani, 2014; Merlis, 2015). We can symbolically write the response of a climate variable Δχ to a CO2 perturbation ΔCO2 as the sum of the temperature-dependent (first term on right-hand side) and direct (second term on right-hand side) changes:
urn:x-wiley:jame:media:jame20928:jame20928-math-0001
with surface temperature Ts. In practice, we use multiple types of GCM simulations to define the direct and temperature-dependent changes, which is described next.

Here, these direct effects are determined by available data from the Coupled Model Intercomparison Project Phase 5 (CMIP5) that supports the United Nations Intergovermental Panel of Climate Change, downloaded using Climate Diagnostic Benchmark query software (CDB query, https://pypi.python.org/pypi/cdb_query/2.0). We use simulation output from several different atmospheric GCMs running the experiments amip4xCO2 and amip, where the former gives an atmospheric response to an immediate quadrupling of CO2 and the latter is its corresponding control simulation. In both experiments, the sea surface temperature (SST) is prescribed to the observed distribution, including interannual variations. Therefore, the resulting differences between the two experiments can be considered these direct effects of increased CO2.

The atmospheric fields respond on fast time scales of the first few months to increased CO2 levels and then stay relatively constant over time in these prescribed SST experiments. Thus, the anomalies are defined as the difference between 30-year climatological means of amip4xCO2 and amip. This is done for each of the seven models listed in Table 1, and the model mean is used to define the perturbations to input fields of the ocean-only GCM. In addition, we isolate the atmospheric fields' impact on OHU by only considering the changes over the ocean in our experimental design and plots. The seven models were chosen based on having complete sets of variables that we analyze for both the amip and coupled simulations (described next).

Table 1. The Ensemble of CMIP5 Models Used in Calculation of the Direct Atmospheric Response to Radiative Forcing
Atmospheric
Model Institution Resolution
CanAM4 Canadian Centre for Climate Modelling and Analysis 2.8°×2.8°
CNRM-CM5 Centre National de Recherches Meteorologiques 1.4°×1.4°
IPSL-CM5A-LR Institut Pierre-Simon Laplace 1.9°×3.8°
IPSL-CM5B-LR Institut Pierre-Simon Laplace 1.9°×3.8°
MIROC5 Japan Agency for Marine-Earth Science and Technology 1.4°×1.4°
MRI-CGCM3 Meteorological Research Institute 1.1°×1.1°
NorESM1-M Norwegian Climate Centre 1.9°×2.5°

We present the following direct atmospheric changes: zonal and meridional surface wind, net precipitation (precipitation minus evaporation), surface downwelling shortwave radiation, and surface downwelling longwave cloud radiative effect. For the purpose of our study, we define “net precipitation” as precipitation minus evaporation, excluding the contribution from runoff, and the “cloud radiative effect” as the difference between all sky and clear sky radiative fluxes. The change in surface shortwave radiation is dominated by cloud radiative effect changes in the AMIP simulations, as the change in clear-sky atmospheric absorption of shortwave radiation is small. Perturbation of the anomalous downwelling longwave cloud radiative effect allows us to consider the longwave radiation's direct response to CO2 forcing while neglecting any changes that would convolve the ocean response with some temperature-dependent changes. These atmospheric fields, along with the surface air temperature and surface air specific humidity, are the set of atmospheric variables used for surface boundary conditions of the ocean GCM simulations.

For comparison, we also look at the coupled response, or total change, of these surface atmospheric fields due to radiative forcing, which is diagnosed by data from coupled GCMs.

We use output from the coupled simulations abrupt4xCO2 and piControl of CMIP5 to analyze the atmosphere's coupled response to radiative forcing. The perturbation experiment, abrupt4xCO2, is forced by an instantaneous quadrupling of CO2 and the control experiment, piControl, is its corresponding control simulation with CO2 and other radiative forcing agent concentrations of the preindustrial era. We take the simulation results from the seven fully coupled versions of the models listed in Table 1 and use the oceanic, model mean values from years 91–100 in analysis.

Figure 1 shows the direct oceanic, zonal mean change in surface winds (Figures 1a and 1c), net precipitation (Figure 1e), downwelling shortwave radiation (Figure 1g), and downwelling longwave cloud radiative effect (Figure 1i) from the amip4xCO2 and amip simulations. These changes are compared to the climatological zonal mean values as well as the coupled oceanic, zonal mean changes between the abrupt4xCO2 (averaged over years 91–100) and piControl simulations in Figures 1b, 1d, 1f, 1h, and 1j. To show the robustness (or lack thereof) in the direct response, the individual models are plotted with thin lines and the ensemble mean is the thick line. Overall, there is both a strengthening and a poleward displacement of surface winds, most noticeably with the Southern Ocean westerly winds, which is robust across models (see also, Ceppi et al., 2018; Grise & Polvani, 2014). The same pattern is seen in both hemisphere's subtropical easterly winds which has a strengthening and poleward shift near the edge of the tropics, and a direct component of the Hadley cell expansion has been previously found (Grise & Polvani, 2014; Merlis, 2015). The Southern Hemisphere has a factor of 2 larger magnitude response than the Northern Hemisphere. For the meridional component of wind (Figures 1c and 1d), the most significant change is the intensification of winds in the midlatitudes for both the Northern and Southern Hemispheres. These meridional surface wind changes are consistent with Ekman balance in the atmosphere's boundary layer.

Details are in the caption following the image
Annual mean and zonal mean direct changes in (a) surface zonal wind, (c) surface meridional wind, (e) net precipitation, (g) surface downwelling shortwave radiation, and (i) surface downwelling longwave cloud radiative effect due to instantaneous quadrupling of CO2 for both the individual models (thin colored) and the model mean (thick black). Zonal mean and ensemble mean direct changes (black solid) and coupled changes from years 91–100 (black dashed) due to instantaneous quadrupling of CO2 and climatological zonal mean values (gray) of the corresponding quantities listed on the left. Note the difference in scale of the y axis between the plots on the left and both y axes on the right.

The coupled GCM changes in both the zonal and meridional surface wind components have a strengthening and poleward shift of winds in the Southern Hemisphere (Figures 1b and 1d), a consistent pattern with the direct changes yet at a larger magnitude. In particular, the Southern Ocean westerlies increase in magnitude by about 30%. The meridional winds are also stronger along the equator, as the southerly and northerly winds north and south of the equator increase in magnitude, respectively. Such a pattern would be consistent with a strengthening of the Hadley Cell, though this is an oceanic mean.

Figure 1e shows that the direct changes to net precipitation are largest in the tropics. Comparing these changes to the climatology, there is a decrease in the region of high tropical precipitation. This is the opposite of the coupled models' “wet-gets-wetter” response pattern (Figure 1f): Regions of climatologically high net precipitation have an increase in net precipitation and regions of climatologically negative (dry) precipitation have a decrease in net precipitation, as a result of temperature-dependent humidity changes (Held & Soden, 2006; Stocker et al., 2013). This indicates that the direct response is much smaller in magnitude relative to the temperature-dependent response to radiative forcing. The values of direct change in net precipitation are consistent with analogous values shown in Bony et al. (2013). Furthermore, the coupled response shows more precipitation along the equator and more evaporation in the subtropics, a pattern consistent with the near-surface convergence associated with the wind changes.

The ensemble mean direct change in downward shortwave radiation, shown in Figure 1g, has a global mean reduction of about 0.35 W/m2. The ensemble mean coupled response of the shortwave radiation (Figure 1h) has a larger magnitude reduction of about 3.5 W/m2. However, these changes lack robustness across models, consistent with previously described top-of-atmosphere cloud radiative effects (Andrews, Gregory, Forster, et al., 2012; Gregory & Webb, 2008). The changes in shortwave radiation of each model neither follow the same zonal pattern nor agree on whether the response to increased CO2 will increase or reduce the amount of solar radiation that reaches the surface. This spread is a manifestation of the uncertainty of the models' parameterization of the clouds.

Figure 1i shows that the downwelling longwave cloud radiative effect is reduced by about 6.8% in direct response to radiative forcing (Figure 1i), which is generally consistent across models. This reduction is the largest in the midlatitudes and minimal across all other latitudes in comparison, a pattern that opposes its climatological state (Figure 1j). The magnitude of direct change is roughly three quarters of the coupled response (Figure 1j), but the coupled response is more prevalent across all latitudes and maximized at the equator and in the subtropics. Interestingly, the coupled response of longwave cloud radiative effect to radiative forcing is near zero in the Southern Ocean, where its direct effect is the strongest.

3 Ocean GCM Simulations

The methods used to investigate how the previously examined direct effects relate to OHU is detailed in this section. We utilize the Geophysical Fluid Dynamics Laboratory ocean GCM, Modular Ocean Model (MOM5) to run simulations perturbing atmospheric input parameters by the direct atmospheric response to CO2. This model uses a 1° horizontal resolution with enhanced (1/3°) latitudinal resolution in the tropics. Its depth resolution is also nonuniform, with 50 levels and coarser resolution (maximum spacing of 371 m) at deeper ocean levels than at sea surface (minimum spacing of 10 m). The sea ice component of the model is the Geophysical Fluid Dynamics Laboratory Sea Ice Simulator and has the same resolution as its ocean counterpart. As our control simulation, we utilize the first of the Coordinated Ocean-ice Reference Experiments, om3_core1, as described in Griffies et al. (2009). The ocean model is forced by input surface atmospheric parameters according to Normal Year Forcing (NYF; Large & Yeager, 2004). These atmospheric fields include surface winds, surface freshwater fluxes of precipitation and runoff, surface downwelling radiative fluxes, surface air temperature, and specific humidity. With the atmospheric input data, the model calculates the surface energy fluxes and mechanical forcing onto the ocean online. The turbulent fluxes are calculated using the bulk formulae (Griffies et al., 2009), where the sensible heat flux is proportional to the difference between air and SST, and the latent heat flux is proportional to the difference between surface air and saturated specific humidity.

In addition to the freshwater fluxes calculated from the prescribed precipitation field, the model applies a salinity-restoring condition (Griffies et al., 2009). To determine the surface radiation, the model receives input data for downwelling shortwave and longwave radiation and computes the net flux of radiation given data for surface albedo and upward longwave flux using the simulated SST. To spin up the model, MOM5 is run with NYF for 500 years to reach an initial quasi-equilibrium.

For each of the atmospheric fields in focus, we apply a perturbation to the input fields according to the direct changes seen from the CMIP5 atmospheric simulation data sets (Figure 1). In doing so, the changes in the ocean's upper boundary condition by the surface variables are altered by the direct effects of radiative forcing on the atmosphere. The perturbation fields are spatially dependent (including variations in longitude) and derived from the monthly climatological anomalies of an instantaneous quadrupling of CO2 from all years of the perturbation. For atmospheric fields that are loaded by MOM5 with a more frequent temporal resolution, the monthly perturbation data are interpolated in time following Killworth (1996) in a manner that does not compromise the monthly mean.

In total, we perform four 100-year-long perturbation simulations. We isolate the perturbations to the meridional and zonal surface winds (wind), net precipitation into the ocean (precip), downwelling shortwave radiation (swrad), and downwelling longwave cloud radiative effect (lwcre). For the downwelling longwave cloud radiative effect simulations, the perturbations to the ocean-only model have the global mean change removed to avoid provoking temperature-dependent changes; however, Figure 1i shows the direct changes to increased atmospheric CO2 including the nonzero global mean. In each perturbation simulation, the surface air temperature and surface air specific humidity fields, used in the model's surface energy budget, are maintained at the control values.

The unchanged surface air variables can lead to offsetting changes in the turbulent surface fluxes, which can reduce the amplitude of the resulting perturbations. For example, a warmer simulated SST should lead to a warmer surface air temperature, all else equal. The air temperature change, which our simulation design neglects, should in turn reduce the magnitude of the restoring of the surface energy budget by turbulent fluxes. Therefore, the magnitude of the responses in these uncoupled simulations could be thought of as a lower bound on the oceanic changes provoked by these atmospheric changes. While this type of coupled air-sea interaction is not captured by the ocean GCM, the representation of the surface energy budget is closer to coupled models than the commonly used thermal relaxation surface boundary condition.

4 Results of Ocean GCM Simulation With Direct CO2 Atmospheric Changes

Figure 2a shows the time series of OHU for the four perturbation simulations. Here, OHU is defined as the time derivative of the change in the global mean total energy between the perturbation and control simulations. The precipitation, shortwave radiation, and longwave cloud radiative effect simulations show little change and thus result in minimal warming or cooling, while the wind simulation has an immediate decrease in OHU of ≈0.3 W/m2 that decays over the 100 simulated years. In other words, the direct atmospheric wind change from increased CO2 leads to ocean heat loss rather than ocean heat uptake. Although the shortwave radiation perturbation induces a cooling tendency on the ocean, it is compensated by offsetting tendencies from the longwave radiation and turbulent fluxes. Given this, the most significant direct impact is that of the surface wind perturbation, which induces a change in the global mean OHU that is included in the temperature-dependent formulation of the two-layer EBM (see section 5). With that said, this change to the global mean energy balance is only a fraction of what would occur in a coupled climate scenario where the warming associated with a quadrupling of atmospheric CO2 would be ≈7 W/m2 (Andrews, Gregory, Webb, et al., 2012).

Details are in the caption following the image
Time series of (a) ocean heat uptake (OHU) and (b) change in maximum strength of the Atlantic Meridional Overturning Circulation (AMOC) at 30°N for each of the four perturbation simulations: surface winds (blue), freshwater flux (green), shortwave radiation (red), and longwave cloud radiative effect (orange).

Figure 2b displays the time series of the change between each simulation and the control in Atlantic Meridional Overturning Circulation (AMOC) strength, Ψ, a value typically used to diagnose the ocean's circulation (Griffies et al., 2009; Rugenstein et al., 2016; Sévellec & Fedorov, 2011). Ψ is defined as the vertical and meridional mass transport in units of Sv=109 kg/s. In our calculations, the total mass transport is a residual mass transport that includes the Eulerian mean transport as well as the parameterization of the eddy-induced velocity (Gent et al., 1995). Following Stocker et al. (2013), we examine the maximum strength of the AMOC along the 30°N latitude band. The direct effect of surface wind to radiative forcing strengthens the overturning circulation by ≈0.5 Sv, and then returns close to the control value by the end of the century. Note that this change in the AMOC is small compared to the steady decline of 4–10 Sv that is expected by the end of the century in RCP8.5, a scenario with comparable radiative forcing (Cheng et al., 2013). The AMOC's response to the surface wind change is larger than that of the simulations with precipitation, shortwave radiation, and longwave cloud radiative effect perturbations (Figure 2b). A strengthening of the overturning in response to increased and poleward shifted westerlies has been found in other GCM simulations pertaining to the Southern Hemisphere (e.g., Jones et al., 2011) and is consistent with existing theories that state that the overturning circulation is determined by the surface wind stress in the Southern Ocean (e.g., Nikurashin & Vallis, 2012). Here, we see a strengthening from the direct wind response of both hemispheres, though the increase in surface winds is much greater in the Southern Hemisphere. We also note that this result is modest relative to the coupled model changes that are likely dominated by temperature-dependent changes.

Since the influence of the direct effects on OHU is dictated by the surface wind perturbations, we focus our analysis on the results from the wind simulation alone. To understand the loss of ocean heat, we examine the surface energy budget. Figure 3 shows the time series of the change in the surface energy budget between the wind perturbation and control simulations. Following the convention in Figure 2a, a positive change is associated with a warming and a negative change indicates a cooling ocean tendency. Figure 3 dissects the total change into each of the major components of the surface energy budget; the radiation contributions (shortwave and longwave) and the turbulent flux contributions (latent energy and sensible heat). We see that the immediate cooling is coming dominantly from the latent heat flux term, which stays negative during the entire century. Note that this wind-induced increase in the latent or evaporative surface flux is opposite in sign to the radiatively driven changes in the surface energy balance, which leads to a decrease in evaporation and has been described in assessments of direct effects or adjustments of the global hydrological cycle (Andrews et al., 2010). Although the latent heat flux term has the strongest influence at the beginning, competing terms like the slowly increasing longwave radiation and sensible heat contribution bring the total change in surface energy budget back to a near equilibrium by the end of the 100 simulated years.

Details are in the caption following the image
Time series of changes in the total surface energy budget (black) between perturbation wind and control simulations, decomposed into the contributions from net shortwave (SW) radiation (red), net longwave (LW) radiation (purple), latent heat flux (green), and sensible heat flux (blue).

Given that the larger release of latent energy is causing the global mean cooling under the perturbation wind simulations, it is necessary to distinguish which aspect of the bulk formula gives rise to this result. Figure 4 displays the change in latent heat flux into the ocean alongside the change in SST and the change in surface wind magnitude for the mean of years 51–60 after imposing perturbation. The regional patterns shown in Figure 4 are the same for the first few years of the perturbation simulation but grow in magnitude over time. The spatial pattern of latent energy cooling is largely related with the pattern of surface temperature change, particularly in the north Pacific and Atlantic regions. The change in latent heat flux has an ambiguous relation to the pattern of change in wind magnitude. From this, we conclude that the larger release of latent energy into the atmosphere is associated with localized regions of warming (discussed more in what follows). Or, more simply, the latent heat flux is more sensitive to the changes in surface temperature than it is to changes in surface wind magnitude. This explains that the response of the latent heat flux term is a result of the regional changes in temperature provoked by ocean circulation changes, rather than the wind speed that enters the bulk aerodynamic formula used to determine the latent energy fluxes. Therefore, we further examine the change in heat content of the ocean resulting from the surface wind changes.

Details are in the caption following the image
Time mean change in (a) sea surface temperature, (b) latent heat flux (positive denotes into the ocean), and (c) magnitude of surface wind between perturbation wind and control simulations for years 51–60.

Figure 5 shows the control residual (Eulerian and parameterized eddy transport) overturning circulation with the zonal mean change in temperature (Figure 5a) and the zonal mean change in circulation (Figure 5b), where the change is the difference in temperature and circulation between the wind perturbation and control simulations for the decade that begins 50 years after the perturbation was applied. In response to the wind, the areas of largest cooling occurs in the top 1,000-m depth. In particular, intense cooling lies right underneath the tropical shallow overturning circulation which is positive north of the equator, and negative southward. Additionally, Figure 5b shows a strengthening of the tropical shallow overturning cells (where the positive circulation is more positive, and the negative circulation is more negative) due to the increase in wind-driven circulation and inducing the cooling as it spreads to greater depths. The strengthening of the shallow overturning cell is larger in the Southern Hemisphere than in the Northern Hemisphere, but each undergoes strengthening from its control state. The pockets of cooling underneath the equator are similar to the findings of England et al. (2014) who discussed an acceleration of the trade winds as the mechanism to increase equatorial upwelling and generate a cooling close to the surface. The overall strengthening of the circulation in the Southern Hemisphere is attributed to the larger wind-driven upwelling that occurs with stronger wind stress shifted southward, as well as an increase in Ekman transport, (e.g., Delworth & Zeng, 2008; Sévellec & Fedorov, 2011). However, these two processes have competing implications for heat content of the Southern Ocean. Ferreira et al. (2015) explains that initially, the enhanced Ekman transport produces a rapid cooling while the increased upwelling south of 55°S gradually warms the Southern Ocean over time as the upwelling occurs where temperature stratification is inverted. In our simulation results, however, we do not notice a strong warming in the Southern Ocean after several decades. The climatological stratification in our model is more isothermal than that of Ferreira et al. (2015), which could explain this discrepancy by limiting the effect of enhanced upwelling on the SST.

Details are in the caption following the image
Zonal mean change in (a) temperature with depth (colored contours) with the strength of the control global residual overturning circulation (black contour lines, intervals of five, solid are positive, dashed are negative) and (b) residual streamfunction overturning circulation (colored contours and black contour lines). Positive values correspond to clockwise circulation, and all values of change are taken as the mean difference between the wind perturbation and control simulations for years 51–60.

This response of the circulation induces the regional warming and cooling displayed in Figure 6a, which shows the change in depth-integrated temperature over the first 1,000 m between the wind forced and control simulations. Outside of the lower latitudes, Figure 6a generally shows the same pattern of warming and cooling as the SST (Figure 4a), just with stronger magnitudes. In particular, the intensity of change in the upper temperature of the Atlantic compared to the SST indicates most of the warming and cooling is occurring under the surface. Geographically, there is approximately uniform cooling along the equator in the Pacific, which is also a feature of the coupled response to a quadrupling of atmospheric CO2. Rugenstein et al. (2016) shows a reduction in SST of ≈0.3 K in that region only within the first couple years of forcing before temperature-dependent processes dominate the response. This also matches the cooling under the tropical overturning cells seen in Figure 5. The warming regions are more localized and intense in areas of strong western boundary currents in the Northern Hemisphere, where there are also large increases in latent cooling (Figure 4). In all, the global mean SST has decreased by ≈0.01 °C by years 51–60. This global mean change is about 0.3% of the total, coupled response in the midcentury (≈3 K) (Geoffroy et al., 2013). In contrast to the global mean SST, the regional SST changes are locally ≈0.5 °C (Figure 4), which is on the order of 10% of the total, coupled response at midcentury.

Details are in the caption following the image
Mean change in (a) vertically integrated temperature for the upper 1,000-m depth and (b) wind stress curl between wind perturbation and control simulations, average difference for years 51–60. Zero wind stress curl for control simulation shown with black line.

Figure 6b displays the change in wind stress curl between the two simulations and gives insight into the spatial pattern of the upper ocean temperature change shown in Figure 6a. The maximum change in wind stress curl of the NH midlatitudes (≈−2.5 × 10−8 N/m3) corresponds to an Ekman pumping of about 2 cm/day. A negative change in wind stress curl describes a stronger northward and westward gradient in wind stress for the perturbation than the control or a weaker southward and eastward gradient. Conversely, a positive change in wind stress curl is given by a stronger southward and eastward gradient or a weaker northward or westward gradient between the two. For instance, in the Southern Ocean, the negative and positive change in wind stress curl just south and north of 60°S, respectively, indicates that the wind stress along 60°S is larger with the perturbed surface winds. Likewise, the negative value of the change in wind stress curl in the northern Atlantic indicates a stronger subpolar gyre in that region. In both the north Atlantic and Pacific Oceans, there is also a poleward shift of the subpolar gyres.

From these results, we ascertain that the spatial pattern of cooling in the tropics is a result of the strengthening of the Ekman suction at the equator. As the wind-driven circulation intensifies, it brings up otherwise cooler waters closer to the surface. In addition, the pattern of warming in the northern midlatitudes is also explained by the redistribution of heat. The combination of stronger easterlies and westerlies intensifies and shifts northward the cyclonic northern gyres (Figure 6b), allowing the warmer waters at the equator to be advected to the northern end of western boundary currents. This behavior is seen at the northern end of both the Gulf Stream and the Kuroshio. While we focus on the decadal response when there is a quasi-equilibrium for gyres and the subtropical cells, there are similar wind-forced changes during the early years of the perturbation simulations.

We note that there is not a strong connection between the intense increase in upwelling and the relatively weak change in temperature along the Antarctic Circumpolar Current in the upper layer of the ocean. This aligns with the findings of Winton, Griffies, et al. (2013) that when the currents are allowed to evolve freely under radiative forcing, a 6-Sv increase in the Southern Ocean Ekman upwelling is associated with only a 0.4-K increase in zonal mean temperature in the ocean interior under radiative forcing. In comparison, our results show a 3-Sv increase in the Southern Ocean Ekman upwelling with a 0.2 °C cooling. While there is an increase in upwelling, the temperature change in that region is modest (Figure 4).

Given the changes in OHU, we examine time series of upper level and deep OHU in Figure 7. The upper level OHU follows the trend of the total ocean, initially cooling then approaching equilibrium, which shows that most of the cooling is in the upper layer. The decomposition of OHU is sensitive to the choice of level, though the results are broadly similar for boundaries between 2,000 and 500 m. The strength of cooling in the upper layer is expected given the vertical structure of the temperature change shown in Figure 5. In addition, Figure 7 reveals that the deep ocean has a slight warming response initially, which is consistent with Rugenstein et al.'s (2016) findings that a near-instantaneous deep warming at higher latitudes occurs under an abrupt quadrupling of atmospheric CO2. This pattern suggests that the intensification and poleward shift of surface winds act to enhance the heat exchange between the surface layer and the deeper ocean by strengthening the global overturning circulation.

Details are in the caption following the image
Time series of ocean heat uptake (OHU, black) due to the direct effect of surface winds to radiative forcing, decomposed into that of the deep layer (dark blue) and that of the upper layer of 1,000 m (light blue).

The ocean GCM simulations show that the direct atmospheric response to CO2 provokes a change in the global mean OHU, meridional overturning circulation, and wind-driven ocean circulation. The changes from the direct response to freshwater fluxes and cloud radiative effects, both shortwave and longwave, are relatively modest compared to those from the surface winds. The vertical partition of the OHU changes forms the basis of our modification of the two-layer energy balance models (EBMs) described in the next section.

5 Direct Responses in Two-Layer EBMs

Beyond comprehensive coupled models of climate change, the ocean's role in determining the magnitude of transient global warming has been encapsulated by two-layer EBMs (Gregory, 2000; Geoffroy et al., 2013; Held et al., 2010). The model represents global mean heat uptake by the shallow and deep layers of the ocean by the following equations:
urn:x-wiley:jame:media:jame20928:jame20928-math-0002(1)
urn:x-wiley:jame:media:jame20928:jame20928-math-0003(2)
The first equation, representing the upper layer of the ocean, models the ocean's response to radiative forcing urn:x-wiley:jame:media:jame20928:jame20928-math-0004 on a faster time scale, where T is the change in global mean surface temperature, λ is the climate feedback parameter, and cF is the heat capacity of the ocean in this top layer. Given that we compare to an ocean GCM, we follow Kostov et al. (2014) and interpret the upper layer as representing the upper ocean, rather than the sum of the upper ocean and atmosphere (e.g., Held et al., 2010). The second equation, where TD and cD are the temperature change and heat capacity for the deep layer, describes the slow response of the deep ocean to radiative forcing via the heat exchange urn:x-wiley:jame:media:jame20928:jame20928-math-0005 between the layers. This is represented by
urn:x-wiley:jame:media:jame20928:jame20928-math-0006(3)
where γ is the OHU efficiency parameter, which governs how sensitive the exchange is to the temperature difference between the layers.

These EBMs can readily emulate comprehensive GCMs' global mean climate response to forcing (Geoffroy et al., 2013; Gregory & Forster, 2008; Held et al., 2010; Kostov et al., 2014; Kuhlbrodt & Gregory, 2012), and parameter values from one radiative forcing scenario can successfully capture the GCM behavior in another (Geoffroy et al., 2013; Merlis et al., 2014). Note that this EBM framework represents OHU as a temperature-dependent process. In other words, this approach that describes OHU as exclusively dependent on surface air temperature and ocean mixing, which is also assumed in diffusive models of, for example, Caldeira and Myhrvold (2013).

Here, we adapt the two-layer EBM to account for the direct effects of the surface winds on the OHU. Using an initial condition of T(0)=0 and TD(0)=0, we set the parameters of the two-layer EBM (equations 1 and 2) as follows: γ=1.6 W·m−2· K−1 following Gregory (2000), cF=4 years and cD=200 years (Held et al., 2010) and λ= W·m−2· K−1, a value that implies an equilibrium climate sensitivity of 3.6 K for a doubling of CO2.

To include the impact of surface winds in this model, a decaying function is applied to the trend of the total ocean's heat uptake (Figure 7) that fits with an amplitude of A=0.27 W/m2 and a decay time scale of τ=48 years (obtained by minimizing the squared error between the ocean GCM results and the exponential function). This function is then added to the equation of heat exchange such that
urn:x-wiley:jame:media:jame20928:jame20928-math-0007(4)

With the additional term, the equation emulates the behavior of the ocean GCM simulations where, due to the surface winds changing near instantaneously, the heat exchange initially reacts to direct effects of forcing, then approaches a regime in which it depends exclusively on the difference in temperature between the two layers.

Analysis of the modified two-layer EBM with a forcing term of 0 W/m2 shows the ocean's response to the direct effects in isolation. Figure 8a shows the time series of the OHU for the shallow and deep ocean according to the two-layer EBM with the heat exchange modification for this case of null radiative forcing. It captures well the immediate cooling of the shallow layer and warming of the deep layer as is found in the MOM GCM simulations (Figure 7).

Details are in the caption following the image
Time series of the ocean heat uptake and global mean temperature according to the two-layer energy balance model in response to the direct effects of surface winds (zero radiative forcing) for both approaches of adaption: (a, c) modifying the heat exchange (equation 4) and (b, d) adjustment to the forcing (equation 6). Both plots display the ocean heat uptake by the top layer of the ocean (red), as well as the deep layer (blue).
Although modifying the heat exchange formulation is a viable representation of the change in ocean circulation (transient strengthening), it is limited in its ability to replicate the total ocean's cooling response (Figure 7). Therefore, it is conceivably more useful to modify the two-layer EBM by incorporating a term that acts as an adjustment to the radiative forcing. We define the forcing of the system urn:x-wiley:jame:media:jame20928:jame20928-math-0008 to be a sum of two terms
urn:x-wiley:jame:media:jame20928:jame20928-math-0009(5)
where urn:x-wiley:jame:media:jame20928:jame20928-math-0010 is the constant radiative forcing and urn:x-wiley:jame:media:jame20928:jame20928-math-0011 is a time-dependent adjustment to the forcing by the direct oceanic effects. This is reasonable if one considers the EBM's upper layer as representing the near-surface ocean (as in Kostov et al., 2014), rather than representing the near-surface ocean and atmosphere (e.g., Held et al., 2010). Therefore, the adjustment to forcing can be described by the decay function fit to the total OHU from Figure 7
urn:x-wiley:jame:media:jame20928:jame20928-math-0012(6)
with the values of A and τ as before. In this scenario, the heat exchange term urn:x-wiley:jame:media:jame20928:jame20928-math-0013 remains proportional to the difference between the temperature of the shallow and deep layers as in equation 3.

Conceptually, the adjustment acts as a direct wind forcing of the ocean that decreases in time as the ocean readjusts to the perturbed wind. This is similar in sentiment to the forcing adjustment presented in Rugenstein et al. (2016), who also uses an exponential function to provide an initial reduction in the radiative forcing that decays over time. Figure 8 shows the impact of this adjustment for the case of null radiative forcing ( urn:x-wiley:jame:media:jame20928:jame20928-math-0014 W/m2). Initially, the shallow layer cools strongly by 0.3 W/m2, then quickly returns to no heat uptake in the first few years. The deep layer cools less significantly at the beginning (≈0.15 W/m2) and then the cooling magnitude gradually increases over 100 years. In this modification, the deep layer has no warming and in fact its behavior is opposite as that of the the ocean GCM simulations (Figure 7); however, it resembles the ocean's total OHU response.

Note that both approaches of modifying the two-layer EBM inadequately preserve the decadal time component of the ocean's response to direct effects. Although cooling happens near instantaneously (Figure 7), it takes a full century to return to equilibrium. This response is intermediate to the two-layer model's fast time scale of 4–5 years but much quicker than the centennial time scales required to reach equilibrium. Some coupled GCMs appear to have a decadal time scale in their surface temperature response to CO2 similar to the intermediate, decadal time scale of our results (Merlis et al., 2014; Winton, Adcroft, et al., 2013). These wind-driven ocean adjustments to the direct surface winds are a possible explanation for that time scale. However, our results suggest that the impact of the direct effects on OHU (0.3 W/m2) are about 4% smaller in magnitude than other processes that occur on fast and centennial time scales and commonly dominate the coupled GCM response to increased atmospheric CO2.

6 Discussion and Conclusions

The transient nature of OHU plays a key role in determining the coupled climate system's response to radiative forcing over time. Therefore, it is a crucial and continuing effort to detail and understand the mechanisms that dictate OHU. In that vein, we examined the impact of atmospheric direct effects of radiative forcing on OHU and circulation. This was done by analyzing how the direct effects, for individual atmospheric fields that affect the ocean's surface boundary conditions, altered the ocean circulation and heat content.

From a focused perspective on the direct atmospheric effects of instantaneous radiative forcing, we find that the surface winds shape the influence of those direct effects on OHU. By dynamically inducing a strengthening of the global meridional overturning circulation, the surface wind perturbations, particularly in the Southern Hemisphere, provide a global mean cooling of the ocean. Such an impact on the ocean circulation implies an initial increase in the heat exchange between the top and deep layers of the ocean and suggests that the direct wind stress forcing mitigates the expected long-term decline of the AMOC (Stocker et al., 2013). This overall cooling of about 0.3 W/m2 happens rapidly and then relaxes toward a new equilibrium. In addition, the maximum impact on circulation occurs around 50 years into the perturbation simulations, which is an intermediate, decadal time scale that does not fit into either a fast time scale of a few years for the near-surface ocean to reach a quasi-equilibrium balance between surface forcing and heat export to the deeper ocean or a slow time scale of centuries for deep ocean equilibriation. To illustrate this influence further, we modify a two-layer EBM to accommodate the cooling provided by the surface winds changes.

We note that there is a discrepancy between our ocean GCM results and those of Rugenstein et al. (2016), whose strengthening of the MOC occurred immediately. This may be the result of cloud radiative effects that vary across models (e.g., Figure 1g).

Although the influence of the direct effects do not determine the centennial trend of warming, they have a transient influence on the decadal time scale that is a cooling tendency that reduces OHU. Given a quadrupling of CO2 has ≈7 W/m2 forcing at the top of atmosphere, the change in OHU by the direct effects is a cooling of 0.3 W/m2, roughly a 5% adjustment. In this way, the OHU response to direct effects can be considered an oceanic adjustment of the global energy balance to radiative forcing (Rugenstein et al., 2016), much like the cloud radiative effect that has been considered a tropospheric, atmospheric adjustment (Andrews, Gregory, Forster, et al., 2012). Here, we assessed the direct effect on the ocean using 4xCO2 simulations in order to examine an idealized scenario with a large signal. Our expectation is that these direct effects on the ocean are also active in more realistic scenarios with transiently increasing radiative forcing (analogous to the documented variations in climate sensitivity, Armour, 2017) or in more modest radiative forcing scenarios, though we have not assessed other scenarios. Therefore, a refined view of the transient nature of OHU should include the impact of the direct effects and their role in limiting OHU at the beginning of radiative forcing scenarios. Furthermore, evaluating the influence of direct effects on OHU in more realistic forcing scenarios would validate that this direct impact is not only present under abrupt forcing. While the temperature-dependent processes likely dominate transient scenarios like the historical forcing over the past 100 years (Stocker et al., 2013), it is possible that the direct effects' strengthening of ocean circulation has mitigated the slow decline in the AMOC over the past century.

Since our study isolated the relative impact of the direct effects on OHU, further work can be done to provide comparison to the temperature-dependent changes and determine how the two components interact and shape OHU collectively. A coupled atmosphere-ocean GCM approach may be better suited to addressing the comparison of direct and temperature-dependent changes.

Acknowledgments

We thank the Natural Sciences and Engineering Research Council (NSERC) of Canada for financial support of this project. CMIP5 simulation results were acquired from the Earth System Grid Federation (ESGF) using the CDB Query, developed by Frédéric Laliberté. We acknowledge the Geophysical Fluid Dynamics Laboratory (GFDL) for the public release of MOM5 and the CORE boundary conditions, Compute Canada for a computing allocation, and helpful discussions with Carolina Dufour and Louis-Philippe Nadeau. All input perturbation and output data from simulations presented here are available at Johns Hopkins University Data Archive (doi:10.7281/T1/PXCQCC).