The Role of Bjerknes and Shortwave Feedbacks in the Tropical Pacific SST Response to Global Warming

The evolution of tropical sea surface temperatures (SSTs) in response to greenhouse warming is of great societal and scientific interest. Most state‐of‐the‐art climate models predict a mean “El Niño‐like” warming pattern by century‐end, characterized by greater warming over the Pacific cold tongue compared to the western warm pool. However, it is unclear which proposed mechanism dominates in this response. Here, we present partially coupled abrupt CO2 doubling experiments in which surface wind stress and shortwave heating are overridden to values from a control simulation. Contrary to previous studies, we find that experiments with overriding of surface wind stress exhibit only 58% of the full reduction in east‐west SST contrast. When both surface wind stress and shortwave flux are overridden, only 34% of the full reduction remains, controlled by spatially‐varying evaporative cooling. These results underscore the importance of Bjerknes and shortwave feedbacks in the tropical Pacific SST response to global warming.

• Climate models predict an eastern equatorial Pacific warming pattern in the tropical Pacific by the end of the 21st century • We use partially coupled climate simulations to assess contributions of wind stress, shortwave, and evaporative feedbacks to this pattern • Weakening of surface wind stress is found to be a major factor leading to a reduced zonal SST gradient over the tropical Pacific Supporting Information: Supporting Information may be found in the online version of this article. 10.1029/2023GL105061 2 of 9 El Niño-like response to CO 2 forcing can be simulated without considering the effect of wind changes.Heede et al. (2020Heede et al. ( , 2021) ) found that in the previous version of the climate model used in this study (the Community Earth System Model version 1; CESM1), with a much weaker projected reduction of the east-west SST gradient, evaporative damping was critical while equatorial wind stress changes played a minimal role for 4 × CO 2 .However, wind changes became much more important for larger CO 2 increases.Perhaps a more sophisticated approach is provided by the partial coupling technique of Lu and Zhao (2012), in which they override the wind stress seen by the ocean component in a coupled climate model.They found, in agreement with the study of Xie et al. (2010), that weakened tropical Pacific winds are not the cause of enhanced eastern equatorial Pacific warming, and that the wind-evaporation-SST feedback dominates their modeled SST response.
We note that debates continue on discrepancies between observed and simulated SST trends over the past century (Seager et al., 2019).While some observational SST data show a slightly steepening east-west Pacific SST gradient over the 20th century (Karnauskas et al., 2009), coupled climate model simulations instead show a weakening of zonal gradients (Coats & Karnauskas, 2017).These discrepancies become especially pronounced over the past 40 years , which show a strong increase in the observed zonal SST gradient and a dramatic strengthening of the Pacific Walker circulation (e.g., Heede & Fedorov, 2023a;Ma & Zhou, 2016;Wills et al., 2022).Several causes for these trends have been proposed, ranging from the ocean dynamical thermostat (Clement et al., 1996;Heede et al., 2020Heede et al., , 2021;;Heede & Fedorov, 2021;Seager & Murtugudde, 1997;Sun & Liu, 1996;Zhang et al., 2019), the effects of anthropogenic aerosols (Heede & Fedorov, 2021), the remote effect of delayed Southern Ocean warming (Dong et al., 2022;Kang et al., 2023), to internal climate variability (Solomon & Newman, 2012).Heede and Fedorov (2023a) argue that some of these factors can work together; for example, observed trends appear to combine a negative Pacific Decadal Oscillation (PDO) pattern and a transient ocean dynamical thermostat.Nevertheless, most global warming projections predict a weakening of the tropical Pacific zonal SST gradient by the end of the 21st century (e.g., DiNezio et al., 2009;Heede & Fedorov, 2021;Li et al., 2023) and there is therefore a need to better understand this response.
In the present study, we expand upon the overriding technique of Lu and Zhao (2012) by conducting a set of simulations using the latest-generation Community Earth System Model version 2 (CESM2).This allows us to quantify the contribution of weakening surface wind stress and changes in shortwave heat flux to the El Niñolike warming pattern simulated by this latest-generation coupled climate model.We conduct several abrupt CO 2 doubling experiments, with and without surface flux override (a technique whereby model-computed surface wind stresses and shortwave heating are overridden with prescribed values).A key innovation compared to Lu and Zhao (2012) is that by also overriding shortwave radiative heating, we can evaluate the contributions of both changes in wind stress and shortwave feedbacks to the reduced zonal SST gradient seen in response to an abrupt CO 2 increase.
The structure of this paper is as follows: in Section 2, we introduce our methodology, including details about CESM2 and our overriding experiments.Section 3 describes the our results, where we assess mechanisms for the SST response to abrupt CO 2 doubling.We conclude and discuss implications of these results in Section 4.

Methods
This study utilizes CESM2.1.3,the latest release of the National Center for Atmospheric Research's (NCAR) coupled general circulation model (GCM; Danabasoglu et al., 2020).CESM2 is a state-of-the-art fully coupled GCM that exhibits high climate sensitivity (Gettelman et al., 2019), along with a prominent reduction in Pacific zonal surface temperature gradients in response to an abrupt CO 2 increase (Heede & Fedorov, 2021).It is therefore a suitable model for investigating the mechanisms for weakened Pacific zonal SST gradients in response to greenhouse warming.CESM2 uses the Community Atmosphere Model version 6 (CAM6) as its atmospheric component, and the Parallel Ocean Program version 2 (POP2) for its ocean component (Danabasoglu et al., 2012).
The ocean timestep and the atmosphere-ocean coupling interval are both 1 hr.We run CESM2 with a nominal horizontal resolution of 1° in the atmosphere and the gx1v7 displaced pole grid for the ocean.For simplicity and numerical stability, the ocean biogeochemistry module is disabled for all experiments; the configuration of our piControl experiment is otherwise unchanged from the default B1850 component set.
First, a preindustrial simulation is initialized from a control simulation spun up for 2,000 years at NCAR.We further spin-up our simulation for an additional 100 years.The final state of this simulation is used as the 10.1029/2023GL105061 3 of 9 initial condition for an experiment where atmospheric CO 2 is abruptly doubled to 569.4 ppm and integrated for 100 years (abrupt2x).Starting from the same initial condition, the piControl simulation is integrated for an additional 100 years, with daily-averages of surface wind stress and surface shortwave heat flux outputted.Next, using saved daily surface wind stress and shortwave heat fluxes, we perform a series of overriding experiments, in which model computed surface wind stresses and shortwave heating are overridden with prescribed values from the piControl to quantify the impacts of changes in surface wind stress and shortwave heating on SST pattern formation in the tropical Pacific.
Several overriding experiments are conducted (Table 1).Starting from the same initial condition as abrupt2x, the τ1c1 experiment involves maintaining atmospheric CO 2 at preindustrial levels, but overriding the model-computed surface wind stresses with daily averages from piControl.The daily stresses are intentionally offset by 1 year from piControl to decorrelate oceanic state variables from the prescribed surface wind stress.This provides a more realistic assessment of overriding-induced mean-state drifts.Similarly, τ1q1c1 involves maintaining atmospheric CO 2 at the preindustrial value, while simultaneously overriding both the model's surface wind stresses and surface shortwave heating (q) with values from piControl.Note that in simulations where τ is overridden, the El Niño Southern Oscillation (ENSO) state is essentially set by wind stresses from piControl and does not evolve freely.This approach is justified by relatively small changes in ENSO in CESM2 under climate change scenarios (Heede & Fedorov, 2023b).Furthermore, exploratory experiments overriding shortwave but not wind stress (q1c1) led to larger mean-state drifts compared to τ1q1c1 (and τ1c1) and unrealistic variability of tropical SSTs; these experiments are therefore not included in our main results.By contrast, simultaneous overriding of τ and q maintains consistency between ENSO and shortwave heating patterns, perhaps explaining the reduced mean-state drift.The SST drifts for τ1c1, τ1q1c1, and q1c1 from piControl are shown in Figure S1 in Supporting Information S1, and the evolution of SSTs over the EEP and WEP is shown in Figure S2 in Supporting Information S1.
The τ1c2 and τ1q1c2 simulations follow the same setup as τ1c1 and τ1q1c1, but involve an abrupt doubling of atmospheric CO 2 .Key differences between this study and previous overriding studies (e.g., Liu et al., 2018;Lu & Zhao, 2012) are that we also override shortwave heating of the surface ocean, we use CESM2 (a high climate sensitivity model which shows a much stronger reduction in the east-west Pacific SST gradient), and we use a much higher horizontal resolution in both the atmosphere and ocean.Furthermore, differencing pairs of experiments provides insight into the role of different mechanisms in the formation of the EEP warming pattern under warming scenarios.Naturally, abrupt2x − piControl reveals the full response of tropical SSTs to greenhouse warming with all plausible mechanisms active (e.g., interactive surface heat fluxes, radiative feedbacks, changes in surface wind stress).With surface wind stresses overriden, the difference τ1c2 − τ1c1 eliminates processes and feedbacks involving the weakening of surface easterly winds as part of the Bjerknes (1969)  and surface winds overridden, isolates the effect of interactive latent/sensible heat fluxes and longwave radiative forcing while eliminating the shortwave effect of clouds.
Finally, by computing differences of differences, we can (to first-order) approximate the contribution of individual mechanisms to SST pattern formation in the tropics.Because τ1c2 − τ1c1 contains all processes except for changes in surface winds, (abrupt2x − piControl) − (τ1c2 − τ1c1) provides an estimate of the role of changing surface wind stress.(τ1c2 − τ1c1) − (τ1q1c2 − τ1q1c1) shows the effect of changing surface shortwave heating, which is expected to primarily reflect the radiative effect of changes in cloud albedo.τ1q1c2 − τ1q1c1 allows for only interactive latent and sensible heat fluxes and longwave radiative effects; the sum of these three terms equals the full response.

Results
Figure 1 shows the formation of patterns of SST anomalies in pairs of abrupt CO 2 doubling experiments.The full response, computed as the difference between abrupt2x and piControl (averaged over years 51-100 and with the domain-mean subtracted) reveals a pattern of prominent relative warming over the Pacific cold tongue (Figure 1a).The SST response to CO 2 doubling when wind stresses are overridden (τ1c2 − τ1c1) still shows enhanced warming over the eastern Pacific, but the pattern is significantly reduced in amplitude (Figure 1b), so that only 58% of the full reduction of the east-west SST gradient remains.This highlights the importance of the Bjerknes feedback for the reduction of zonal SST gradients in CESM2.In this feedback the weakening of the zonal SST gradient leads to the weakening of the zonal wind stress and a further weakening of the SST gradient.When both surface shortwave flux and wind stress are overridden (τ1q1c2 − τ1q1c1), the model simulates a reduction in zonal gradient that is weaker still compared to the full response, and with a slightly different pattern (Figure 1c) compared to τ1c2 − τ1c1.Only 34% of the full reduction of the east-west SST gradient remains, which supports the conclusion of previous studies that cloud radiative effects may play an important role in the reduction of east-west zonal gradients over the tropical Pacific (Burls & Fedorov, 2014;Erfani & Burls, 2019).
The remaining response (Figure 1c) is predominantly controlled by evaporative damping (Knutson & Manabe, 1995) and longwave radiative feedbacks, since surface wind stress and surface shortwave effects (e.g., due to changes in cloud cover) are overridden to piControl.Changes in latent heat flux are expected to dominate in this experiment, following the wind-evaporation-SST (WES) feedback (Knutson & Manabe, 1995;Xie et al., 2010), partially compensated by changes in net longwave fluxes.Changes in sensible heat fluxes are small and play a minimal role.
We find that the change in surface net longwave flux is similar between the warm pool and cold tongue (Figure 2b) and we therefore conclude τ1q1c2 − τ1q1c1 primarily reflects the effects of evaporative damping.Indeed, Figure 1c shows enhanced warming over the cold tongue even with τ and q fixed, although this can only partially explain the full response which is much larger in amplitude (Figure 1a).The difference between τ1q1c2 and τ1q1c1 for surface latent heat flux, net surface longwave, and their sum are shown in Figure 2; other related fields are shown in Figures S3 and S4 in Supporting Information S1. Figure 2 confirms that it is spatially varying evaporative cooling along the equator that is largely responsible for the reduction in the zonal SST gradient, if wind stress and shortwave feedbacks are eliminated.
While the weaker relative warming over the EEP seen in both the τ1c2 and τ1q1c2 experiments compared to the full response indicates that changes in surface wind stress play a major role in the formation of the EEP warming pattern, a further linear decomposition allows us to explicitly estimate the role of each of these individual processes.Because τ1c2 − τ1c1 contains all processes except for changes in easterly winds, (abrupt2x − piControl) − (τ1c2 − τ1c1) isolates the role of changing surface wind stress and is shown in Figure 3b.Although weaker in amplitude, the resemblance of Figure 3b to the full response shown in Figure 1a confirms that changes in τ play a crucial role in the EEP-enhanced warming response.
Next, we directly evaluate the role of changes in shortwave heating in the formation of equatorial Pacific SST patterns.This is computed as (τ1c2 − τ1c1) − (τ1q1c2 − τ1q1c1), isolating the effect of changes in shortwave heating, such as those that may arise from changes in cloud albedo.Indeed, this forcing factor is found to play an important role in the reduction of zonal gradients in CESM2, contributing weakly to relative warming west of 120°W, but leading to a strong warming signal east of 120°W along the equator (Figure 3c). Figure S5 in Supporting Information S1 follows Figure 1, but shows changes in SST without subtraction of the mean warming.
Finally, we show timeseries of the evolution of the east-west SST contrast over the equatorial Pacific for all experiments in Figure 4a, taken as the difference between SSTs over the Western Equatorial Pacific (WEP; 120°E-150°E, 5°S-5°N) and EEP (205°E-280°E, 5°S-5°N).A Gaussian low-pass filter (σ = 10) has been applied to all curves to isolate the long term trend.The relatively small difference between the τ1c1 and τ1q1c1 simulations compared to piControl confirms that our overriding technique maintains the climatological mean state and does not lead to large climate drifts.The 2 × CO 2 experiment shows an east-west zonal SST contrast that immediately weakens in response to CO 2 doubling and continues to weaken to year 100, not exhibiting the transient steepening argued by Seager et al. (2019) and demonstrated for a number of climate models by Heede et al. (2020Heede et al. ( , 2021) ) and Heede and Fedorov (2021).Figure 4b shows differences in SSTs over the WEP, EEP, and the east-west SST contrast for pairs of simulations, averaged over the final 50 years of the simulations.Consistent with Figure 1, abrupt CO 2 doubling with τ and q overridden leads to a smaller reduction in east-west contrast compared to the full response.

Discussion and Conclusions
In this study, we used a partial coupling technique to investigate factors leading to the reduced east-west SST contrast simulated by CESM2 in response to an abrupt doubling of CO 2 .CESM2 is a model that shows a particularly strong reduction in Pacific zonal surface temperature gradients in response to an abrupt CO 2 increase, and is a suitable model to use for studying these changes.In contrast to previous studies, which suggested that changes in surface wind stress do not play an important role in the emergence of a reduced zonal SST gradient, we found that the weakening of surface winds in fact plays a major role within CESM2.Indeed, our results underscore the potential importance of the Bjerknes feedback in SST pattern formation under global warming scenarios, and suggest that mixed-layer models may be inadequate for understanding projected east-west zonal gradient changes in the tropical Pacific.Furthermore, by overriding shortwave heating, we show a contribution by clouds to the warming signal seen over the eastern tropical Pacific, particularly to the east of 120°W.
We note that the linear decomposition technique used to isolate the effects of τ and q can provide only a first-order estimate of the importance of these mechanisms.We avoid making explicit claims about the relative importance of each of the mechanisms we tested.For instance, while at face value, the effect of shortwave heating (q) accounts for only 24% of the full response (less than the effect of τ), we do not claim that the weakening of τ is more important compared to changes in shortwave radiation because overriding of q is applied to experiments with overridden τ, which by itself leads to a less pronounced reduction in zonal Pacific gradients compared to the full response.Our main conclusion is that changing surface wind stress and shortwave radiative effects both significantly contribute to the weakening of zonal Pacific SST gradients simulated in response to abrupt CO 2 doubling within CESM2.
Finally, we note that while shortwave forcing and weakening trade winds are found to be important mechanisms for the reduced zonal SST contrast in this model, it does not suggest that evaporative damping is unimportant.The residual, when τ and q are fixed, still accounts for around 34% of the full reduction in zonal SST contrast in our simulations.This residual is predominantly controlled by stronger evaporative cooling in the western equatorial Pacific compared to the east.Furthermore, our model experiments cannot distinguish between a weakening of surface winds that is primarily driven by the slowdown of the large-scale atmospheric overturning circulation following thermodynamic arguments (Held & Soden, 2006), or the weakening of winds that results from a Bjerknes-feedback response to SST changes induced by evaporative damping or cloud feedbacks.While it is difficult to separate various contributions to the weakening of surface winds, the magnitude and implications of the thermodynamically driven and feedback-induced changes in surface wind stress should be more closely evaluated in future studies.
Several outstanding questions remain, such as the extent to which these results are model dependent.One plausible explanation for why our results differ from prior studies is that the amplifying effect of surface winds may become important only for a sufficiently pronounced weakening of the zonal SST gradient.For example, in CESM1 (Heede et al., 2020(Heede et al., , 2021)), a weakening of zonal SST gradients comparable to our CESM2 CO 2 doubling simulations is achieved only with abrupt increases in CO 2 of 8× or even 16×, and indeed under such conditions wind changes become important in that model.Prior overriding studies investigated the effects of CO 2 increases in models with relatively low climate sensitivity comparable to CESM1 (Kiehl et al., 2006;Lu & Zhao, 2012), which typically imply modest changes in the Walker circulation and zonal SST gradients (Li et al., 2023).Another issue concerns the timescales of the establishment of the EEP warming pattern.Previous studies suggest that some CMIP6 models exbibit a transient ocean thermostat-like effect in response to a CO 2 increase that may last 20-80 years; only then does the EEP pattern emerge (Heede et al., 2021;Heede & Fedorov, 2021).No such transient thermostat is evident in CESM2 (Figure 4), nor in many other GCMs.Therefore, applying our overriding technique across a range of models, with and without a transient ocean dynamical thermostat could be useful.

Figure 1 .
Figure 1.SST warming patterns in fully and partially coupled abrupt CO 2 doubling experiments, averaged over years 51-100.(a) SST difference between abrupt2x and piControl, showing the full SST response.(b) Difference between τ1c2 and τ1c1 (wind stress overridden to piControl).(c) Difference between τ1q1c2 and τ1q1c1 (both wind stress and shortwave heating overridden to piControl).To highlight patterns of change, anomalies are computed relative to respective baseline experiments with domain-mean ocean warming (60°E-60°W, 40°S-40°N) subtracted.In all panels, black boxes show the domains used for computing the east-west zonal SST gradient.

Figure 2 .
Figure 2. Difference in surface latent heat fluxes and net surface longwave in simulations with wind stress and shortwave prescribed, averaged over years 51-100.(a) Difference in surface latent heat flux between τ1q1c2 and τ1q1c1.(b) As in (a), but for differences in net surface longwave fluxes.(c) The sum of panels (a) and (b).In all panels, black boxes show the domains used for computing the east-west zonal SST gradient.

Figure 4 .
Figure 4. Evolution of tropical Pacific SST gradients in different experiments.(a) Smoothed timeseries of the annual-mean east-west zonal SST contrast in the equatorial Pacific, taken as the difference between SSTs over the WEP (120°-150°E, 5°S-5°N) and eastern equatorial Pacific (EEP) (205°-280°E, 5°S-5°N) in fully and partially coupled abrupt CO 2 doubling experiments and in the piControl.The mean piControl value of 4.1°C has been subtracted, and a Gaussian low-pass filter (σ = 10) has been applied to all curves.(b) Bar chart illustrating differences in SSTs over the WEP, EEP, and the east-west SST contrast averaged over the final 50 years of the experiments.
In the names of overriding experiments, τ refers to surface wind stress, q to shortwave fluxes, and c to CO 2 concentration.1×* indicates that the values are overridden to the piControl, 2× indicates a doubling of CO 2 .-Values computed prognostically from the model (not overriden).*Overridden with daily values from piControl, intentionally offset by 1 year.
feedback, and isolates latent/sensible heat fluxes and radiative effects.The difference τ1q1c2 − τ1q1c1, with both shortwave heating