Tropical cloud-radiative changes contribute to robust climate change-induced jet exit strengthening over Europe during boreal winter

The North Atlantic jet stream is projected to extend eastward towards Europe in boreal winter in response to climate change. We show that this response is robust across a hierarchy of climate models and climate change scenarios. We further show that cloud-radiative changes contribute robustly to the eastward extension of the jet stream in three atmosphere models, but lead to model uncertainties in the jet stream response over the North Atlantic. The magnitude of the cloud contribution depends on the model, consistent with differences in the magnitude of changes in upper-tropospheric cloud-radiative heating. We further study the role of regional cloud changes in one of the three atmosphere models, i.e. the ICON model. Tropical cloud-radiative changes dominate the cloud impact on the eastward extension of the jet stream in ICON. Cloud-radiative changes over the Indian Ocean, western tropical Pacific, and eastern tropical Pacific contribute to this response, while tropical Atlantic cloud changes have a minor impact. Our results highlight the importance of upper-tropospheric tropical clouds for the regional circulation response to climate change over the North Atlantic-European region and uncertainty therein.


Introduction
The North Atlantic eddy-driven jet stream is expected to undergo substantial changes in response to climate change. Climate models project that the annual-mean jet stream will shift poleward (e.g. Chang et al 2012, Barnes andPolvani 2013, Vallis et al 2015), and reanalyses indicate that the vertical wind shear will increase due to changes in meridional temperature gradients (Lee et al 2019). However, the jet response varies strongly between seasons. While a poleward jet shift is found during most seasons, the jet is projected to extend eastward towards Europe rather than to shift poleward during boreal winter (December-February, DJF) (e.g. Pinto et al 2007, Woollings and Blackburn 2012, Zappa et al 2013, Simpson et al 2014, Harvey et al 2015. As shown in Harvey et al (2020), this wintertime response is found in the model-mean of coupled climate models that contributed to phases 3, 5, and 6 of the Coupled Model Intercomparison Project (CMIP; Meehl et al 2000, Taylor et al 2012, Eyring et al 2016. The eastward extension is robust across coupled climate models (Simpson et al 2014) but its magnitude remains uncertain (Shepherd 2014). Over the North Atlantic, the response is uncertain as some models exhibit a poleward jet shift while others exhibit an equatorward jet shift (Barnes andPolvani 2013, Shepherd 2014).
The eastward extension of the North Atlantic jet stream in response to climate change co-occurs with an eastward extension of the North Atlantic storm track (Harvey et al 2020). The responses of the jet stream and storm track are of large social and economic interest, with both positive and negative consequences for Europe. On the one hand, the increases in wind speed will result in a higher wind energy production over Northern Europe (Hueging et al 2013, Reyers et al 2016, Carvalho et al 2017, Moemken et al 2018. On the other hand, an increase in winter storms over Europe will increase the potential for severe losses due to storminess, flooding after extreme precipitation events, and other damages , Pinto et al 2012, Catto et al 2019. Changes in cloud-radiative properties affect the zonal wind response to climate change as clouds and the atmospheric circulation are strongly coupled via radiation (see review by Voigt et al 2021, and references therein). This cloud-radiative impact acts via changes in the surface energy balance and changes in the atmospheric energy balance, referred to as surface pathway and atmospheric pathway of the cloudradiative impact, respectively .
Here, we focus on the atmospheric pathway of the cloud-radiative impact. The atmospheric pathway can be quantified by using the cloud-locking method together with prescribed sea-surface temperatures (SSTs). Prescribing SSTs disables the surface pathway, as then cloud-induced changes in the surface energy balance over the ocean no longer affect SSTs. As a result, the circulation response can be decomposed into contributions from changes in cloud-radiative properties and SSTs (e.g. Voigt and Shaw 2015. The atmospheric pathway of the cloud-radiative impact contributes substantially to the zonal wind and jet stream responses in atmosphere models in the zonal-mean perspective  and across seasons and regions (Albern et al , 2020. In particular, Albern et al (2019) showed for the ICON model that about one quarter of the DJF zonal wind response at 850 hPa across the midlatitudes can be attributed to changes in cloudradiative properties. Further, Albern et al (2020) showed that tropical cloud-radiative changes dominate the cloud impact on the zonal wind response in the same model. Yet, while the zonal-mean response was studied in several models, the impact of cloudradiative changes on the regional zonal wind and jet responses has so far only been quantified in the ICON model.
Here, we study the role of cloud-radiative changes on the eastward extension of the North Atlantic jet stream towards Europe under climate change. We first investigate a hierarchy of climate models and simulation setups to identify which aspects of the climate change response are robust. We then study the impact of cloud-radiative changes on the zonal wind response in three atmosphere models, and identify how much of the robust response can be attributed to cloud-radiative changes in each model. Finally, we focus on the ICON model to assess which regional cloud-radiative changes are most important for the zonal wind response over Europe.

CMIP5 simulations
We investigate the zonal wind response to climate change across models and climate change scenarios of varying complexity. The most complex models in our model hierarchy are coupled climate models. We study the historical (years 1975-2004) and RCP8.5 simulations (years 2070-2099) from 37 coupled climate models that participated in CMIP5 (Taylor et al 2012). Reducing the models' complexity, we further investigate output from eleven atmosphere-only climate models with prescribed SSTs and sea ice cover that performed the Amip, Amip4K and AmipFuture simulations (years 1979-2008) of CMIP5 (Taylor et al 2012). In these simulations, climate change is mimicked by increasing SSTs. The Amip4K climate change scenario is the most idealized scenario in our hierarchy as it simulates climate change by a uniform 4 K SST increase. The AmipFuture simulations, in contrast, use an SST pattern derived from coupled climate models (Taylor et al 2009(Taylor et al , 2012. The investigated CMIP5 models are listed in table S1 (available online at stacks.iop.org/ERL/16/084041/mmedia).

Cloud-locking simulations
We investigate simulations with the atmospheric components of the ICON model (Zängl et al 2015), and the low resolution versions of the MPI-ESM , Stevens et al 2013 and IPSL-CM5A (Dufresne et al 2013) models that applied the cloud-locking (ICON) or cloud-and water vaporlocking (MPI-ESM and IPSL-CM5A) methods to determine how much of the zonal wind response can be attributed to changes in cloud-radiative properties. The ICON simulations with locked clouds and interactive water vapor are taken from Albern et al (2019). The MPI-ESM and IPSL-CM5A simulations with locked clouds and locked water vapor are taken from . The simulations were performed analogously to the Amip simulations, but use climatological SSTs and sea ice cover. They have a length of 27 years (IPSL-CM5A), 28 years (MPI-ESM), and 30 years (ICON), respectively. For each simulation, the first year is excluded from the analysis to avoid effects from model initialization. In accordance with the Amip4K simulations, climate change was mimicked by a uniform 4 K SST increase (see Albern et al 2019 and Voigt et al 2019 for details of the simulations' setups). Detailed descriptions of the locking method are given, for example, in Voigt and Shaw (2015) and Albern et al (2019).
For the cloud-locking method, first the radiative properties of clouds have to be stored for the present-day and climate-change simulations. Second, four simulations have to be performed, in which SST (T) and cloud-radiative properties (C) are prescribed to either of the two climate states. The total locked response of any given variable X is then where the indices indicate whether T and C are taken from the present-day (1) or climate-change (2) simulation. The cloud-radiative impact via the atmospheric pathway is calculated as (Albern et al 2019) Analogously, the radiative properties of clouds and water vapor have to be stored for the cloud-and water vapor-locking method, and eight simulations have to be performed, in which T, C, and water vaporradiative properties (W) are prescribed to either of the two climate states. The total locked response for simulations with prescribed clouds and water vapor is and the cloud-radiative impact via the atmospheric pathway is calculated as (Voigt and Shaw 2015) Note that for all investigated models the residuals between the total response with interactive clouds/ water vapor and the total response with locked clouds/water vapor, which arise due to the decoupling of clouds/water vapor and the circulation when applying the locking methods, were found to be small .
It is meaningful to directly compare the cloudradiative impact from ICON simulations with interactive water vapor to that from MPI-ESM and IPSL-CM5A simulations with locked water vapor because the cloud-radiative impact is largely insensitive to the treatment of water vapor . Investigating the annual-mean zonal-mean atmospheric circulation,  showed for ICON that the estimated cloud-radiative impact on the responses of various circulation metrics, including the position and strength of the jet stream, hardly depends on whether water vapor is interactive or prescribed. Investigating the regional zonal wind response at 850 hPa, ∆u 850 , we find that the treatment of water vapor in the ICON simulations of Voigt and Albern (2019) has a negligible effect on the pattern and magnitude of the total zonal wind response and the cloud-radiative impact on ∆u 850 over the North Atlantic-European region during winter (figure S1).
For the ICON model, we do not only determine the impact of global cloud changes but also the impact of regional cloud changes. In addition to the four above mentioned simulations for the global cloud impact, four more simulations are performed for each region of interest (Albern et al 2020). In these simulations, clouds in the region of interest (marked by subscript a in equation (5)) and clouds in the rest of the world (marked by subscript b) are prescribed to values from either the control simulation or the climatechange simulation. A more detailed discussion of the methodology can be found in Albern et al (2020).
Based on these simulations, the impact of regional cloud changes is calculated as We investigate the regional cloud impacts for the following regions: tropics (30 • S-30 • N, all longitudes), midlatitudes (30 • N-60 • N and 30 • S-60 • S, all longitudes), polar regions (poleward of 60 • N/S, all longitudes), North Atlantic-European for a schematic of the regions).

Jet stream
We derive the eddy-driven jet stream from the maximum in the zonal wind at 850 hPa. Based on the zonal wind interpolated linearly onto a 0.01 • latitude grid, we perform a quadratic fit around the maximum and the two neighboring grid points, and define the jet latitude φ jet and jet strength u jet as the position and value of the maximum of the quadratic fit (e.g. Barnes andPolvani 2013, Albern et al 2019).

Robust circulation response and contribution of global cloud-radiative changes
We begin by showing which aspects of the circulation response to climate change over the North Atlantic-European region are robust across coupled and atmosphere-only climate models. The top row of figure 1 shows the CMIP5 model-mean zonal wind response at 850 hPa, ∆u 850 . In the model mean, all three scenarios show a poleward shift and strengthening of the jet stream over the North Atlantic, and a zonal wind increase over central and northern Europe (figures 1(a)-(c)). The latter is associated with an eastward extension of the North Atlantic jet stream towards Europe, and commonly referred to as jet exit strengthening. The responses over Europe are robust across models in all three model setups. Over the North Atlantic, however, the models do not agree on the u 850 increase on the poleward flank of the jet in the coupled models, and on the u 850 weakening on the equatorward flank of the jet in the atmosphere-only models.
As the CMIP5 model mean, ICON, MPI-ESM, and IPSL-CM5A show the jet exit strengthening over Europe (figures 1(d)-(f)). However, the region of the jet exit strengthening is model dependent. While the zonal wind increase in MPI-ESM, IPSL-CM5A, and the CMIP5 simulations is strongest over western to central Europe, the zonal wind increase in ICON is largest over the southern half of northern Europe including the North Sea and Baltic Sea regions. The region of the largest zonal wind increase is linked to the tilt of the North Atlantic jet stream, which is larger in ICON and smaller in the other two models and the CMIP5 model mean (cf thick black dots in figure 1).
ICON, MPI-ESM and IPSL-CM5A reflect the CMIP5 model uncertainties over the North Atlantic. ICON shows a poleward jet shift across the North Atlantic, while MPI-ESM and IPSL-CM5A exhibit a jet strengthening over the eastern part of the North Atlantic, and in IPSL-CM5A the jet shifts equatorward over the eastern North Atlantic close to France and the Iberian Peninsula (figures 1(d)-(f)). The responses in all three models agree well with the robust zonal wind responses in the Amip4K model mean (hatching in figures 1(d)-(f)). Figure 2 contrasts the jet response over Europe (0 • -25 • E, panels (a)-(c)) with the jet response over the North Atlantic (60 • W-0 • , panels (d)-(f)) across the CMIP5 models and ICON, MPI-ESM, and IPSL-CM5A. In both regions, most models exhibit poleward jet shifts of up to 2.5 • . Several models exhibit an equatorward jet shift over the North Atlantic which is less pronounced over Europe. Some models (CMCC-CMS and CSIRO-Mk3-6-0 for RCP8.5; bcc-csm1-1, IPSL-CM5B and MIROC5 for AmipFuture and Amip4K) exhibit very large jet shifts of more than 10 • . These large jet shifts are excluded from figure 2, and are due to the fact that the models exhibit very weak jet streams over Europe, resulting in a weak and flat u 850 profile that is very sensitive to small wind changes.
While the magnitudes of the jet shifts are similar in both regions, larger differences between the North Atlantic and Europe are found for the jet strength response. In the atmosphere-only models, the jet strengthening over Europe is in most models two to five times larger than over the North Atlantic. The same general behavior is found for the coupled climate models. Yet, several coupled models exhibit only small responses in the jet strength over Europe, reflecting the larger inter-model variability in the more complex coupled models (although this is also partly due to the larger ensemble). In both regions, the jet shift and jet strengthening in ICON, MPI-ESM and IPSL-CM5A lie well within the jet responses of the atmosphere-only CMIP5 models for the Amip4K scenario (figures 2(c) and (f)).
We now focus on the jet exit strengthening over Europe. Figure 3 shows the total u 850 response (reproduced from figure 1) and the cloud impact on the u 850 response in ICON, MPI-ESM, and IPSL-CM5A. The cloud-radiative impact contributes substantially to   the jet exit strengthening in all three models (figure 3, right). Over the North Atlantic, however, the cloud impact differs between the three models so that it can be considered as one source of uncertainty in the circulation response in this region. This finding is consistent with the non-robust circulation response over the North Atlantic in the CMIP5 models as well as in ICON, MPI-ESM and IPSL-CM5A (cf figure 1).
Even though cloud changes appear to robustly contribute to the jet exit strengthening, the magnitude of the cloud impact varies strongly between the three models, as does the total response (figure 3). Further, the relative contribution of the cloud impact to the total u 850 response is model dependent. Over the European region, for which the signs of the total responses in ICON, MPI-ESM and IPSL-CM5A agree with the sign of the robust response of the CMIP5 models in the Amip4K scenario (50 • N-59 • N, 4 • W-25 • E, cf hatching in figures 1(d)-(f)), cloud changes contribute about one quarter to the total u 850 response in ICON and MPI-ESM. In IPSL-CM5A, however, essentially all of the total response in this region can be attributed to cloud-radiative changes. Note that for large parts of the North Atlantic-European region, the pattern of the u 850 response to cloud changes largely resembles the pattern of the total response in ICON and MPI-ESM. In IPSL-CM5A, in contrast, the cloud impact and total response exhibit quite different spatial structures with an equatorward jet shift and jet strengthening for the total response and a poleward jet shift and jet strengthening for the cloud impact.
To understand the different magnitudes and relative contributions of the cloud-radiative impacts in the three models, we investigate the changes in cloudradiative heating derived from partial-radiative perturbation (PRP) calculations (Wetherald and Manabe 1988, Colman and McAvaney 1997, Voigt and Shaw 2016. The PRP calculations are based on the locked simulations and quantify the changes in temperature tendencies due to changes in cloud-radiative properties under climate change. In the zonal mean, the largest changes in atmospheric cloud-radiative heating are found in the tropical and midlatitude upper troposphere (figures 4(a)-(c)). These changes are strongly linked to changes in cloud cover (Voigt and Shaw 2016, Albern et al 2020, and differences in cloudradiative heating changes between the models can be linked to differences in present-day cloud cover and in cloud cover response to climate change (figure S3). For a direct comparison of cloud-radiative heating changes and cloud cover response cf figures 5(g)-(i) in .
Previous studies proposed that changes in highlevel ice clouds play an important role for the response of the midlatitude circulation to climate change (Voigt and Shaw 2016, Albern et al 2020. Thus, we focus our analysis on the upper troposphere and investigate regional vertical-mean changes in atmospheric cloud-radiative heating for a 200 hPa thick layer below the DJF tropopause. The qualitative differences in the magnitude and pattern of the change in atmospheric cloud-radiative heating between the models is independent of whether the vertical mean is calculated over a 200 or 300 hPa thick layer below the tropopause. In all three models, the changes in uppertropospheric cloud-radiative heating peak over the western tropical Pacific and Maritime Continent (figures 4(d)-(f)). In ICON and MPI-ESM, there are secondary peaks over the Indian Ocean, while in IPSL-CM5A a secondary peak is found over the central subtropical Pacific of the Southern Hemisphere. The changes in atmospheric cloud-radiative heating in ICON and MPI-ESM are similar in a sense that they are largest in similar tropical regions, while the changes in the midlatitudes and polar regions are small (figures 4(d) and (e)). This might explain why the relative contributions of the cloud impacts on the u 850 response in ICON and MPI-ESM are similar. In IPSL-CM5A, the peak in the vertical-mean tropical upper-tropospheric cloud-radiative heating changes is smaller while the changes in the midlatitudes are larger than in ICON and MPI-ESM (figure 4(f)). The increased cloud-radiative heating around the jet stream might explain the larger cloud impact on the u 850 response in IPSL-CM5A compared to the other two models.
The results suggest that differences in the pattern and magnitude of the upper-tropospheric cloudradiative heating changes can lead to differences in the u 850 response in ICON, MPI-ESM and IPSL-CM5A (cf figure 3). As Albern et al (2020) showed that tropical cloud-radiative changes dominate the u 850 response to climate change in ICON, the differences in the u 850 response might be primarily linked to differences in tropical cloud-radiative heating changes. Therefore, we investigate the impact of tropical cloudradiative changes in more detail in the next section.

Regional cloud-radiative impact on the circulation response in ICON
In this section, we focus on the ICON model to investigate which regional cloud-radiative changes are most important for the global cloud impact. Albern et al (2020) showed that tropical cloud-radiative changes dominate the annual-mean, wintertime and summertime global cloud-radiative impact on the midlatitude u 850 response to climate change in ICON (cf their figure 3). Here, we investigate the wintertime u 850 response over the North Atlantic-European region in more detail. We find that tropical cloud-radiative  figure 5), tropical cloud changes actually lead to a larger zonal wind increase than global cloud changes, while about one fifth and one quarter of the jet exit strengthening can be attributed to midlatitude and polar cloud changes, respectively. Note that the sum of the tropical, midlatitude and polar cloud changes overestimates the global cloud impact in the given region by more than 50% due to non-linearities that arise when the cloudradiative heating is induced individually (Butler et al 2010), an effect that might be enhanced by gradients in the cloud-radiative properties at the boundaries of the tropical, midlatitude and polar regions (Albern et al 2020).
Over the North Atlantic, tropical, midlatitude and polar cloud changes all contribute to the poleward jet shift, and tropical cloud changes lead to a In contrast, local cloud-radiative changes over the North Atlantic and Europe lead to a slight, non-significant weakening of the zonal wind and jet stream over the North Atlantic-European region (not shown) (cf figures S2(a)-(d) for the regions). Thus, remote cloud-radiative changes, in particular those in the tropics, are much more important for the jet stream response over the North Atlantic-European region than local cloud-radiative changes.
We now investigate which tropical region dominates the tropical cloud-radiative impact. Cloud changes over the western tropical Pacific (WP), the eastern tropical Pacific (EP) and the Indian Ocean (IO) (cf boxes in figures 4(d) and S2) all contribute to the jet exit strengthening over Europe (figures 6(a)-(c)). In the region with the strongest zonal wind increase (boxes in figure 6), the areamean tropical cloud impact (1 m s −1 ) is dominated by EP cloud changes (0.45 m s −1 ) followed by WP (0.36 m s −1 ) and IO (0.23 m s −1 ) cloud changes. In contrast, the impact of tropical Atlantic (TA) cloud changes (0.07 m s −1 ) is small ( figure 6(d)). Over the North Atlantic, EP and TA cloud changes are most important for the pattern of the u 850 response and the jet strengthening, but all four tropical regions contribute to the poleward jet shift of the North Atlantic jet stream. Note that while dividing the global cloud impact into tropical, midlatitude and polar cloud changes results in a substantial overestimation of the global cloud impact, dividing the tropical cloud impact into WP, EP, IO and TA cloud changes results in a comparably weak overestimation of the tropical cloud impact across the North Atlantic-European region (figures 6(e) and (f)). The overestimation in the region of the jet exit strengthening amounts to only about 12.5%.
Our results show that no smaller tropical region dominates the tropical cloud-radiative impact on the jet exit strengthening. This result is independent of whether we investigate the sum of the individual cloud impacts or the cloud impact that results from simultaneous cloud changes in the different tropical regions (not shown), and indicates that large-scale processes and interactions, such as the Walker circulation, are important for the circulation response over Europe. Further, the change in atmospheric cloud-radiative heating has a rather complex spatial structure, making it difficult to select smaller regions without introducing heating gradients that might affect the circulation response to tropical cloudradiative heating.

Conclusions
We investigated the atmospheric pathway of the cloud-radiative impact on the zonal wind and jet stream responses to climate change over the North Atlantic-European region during boreal winter. The jet exit strengthening, i.e. the eastward extension of the North Atlantic jet stream towards Europe and the associated zonal wind increase over Europe, is robust across coupled and atmosphere-only climate models and climate change scenarios. At the same time, the zonal wind response over the North Atlantic is not robust. Global cloud-radiative changes contribute robustly to the jet exit strengthening in simulations with the atmospheric components of ICON, MPI-ESM and IPSL-CM5A that apply the cloudor cloud-and water vapor-locking methods. Further, cloud-radiative heating can be considered as one source of model uncertainty in the zonal wind and jet stream responses over the North Atlantic. Differences in the absolute and relative contributions of the cloud impacts are related to differences in the magnitude and pattern in the upper-tropospheric change in atmospheric cloud-radiative heating in the three models.
Tropical clouds dominate the cloud-radiative impact on the jet exit strengthening in ICON. Indian Ocean, western tropical Pacific and eastern tropical Pacific cloud changes all contribute to the jet exit strengthening while tropical Atlantic cloud changes have a minor impact. This is consistent with the changes in atmospheric cloud-radiative heating, which are largest over the tropical Pacific and Indian Ocean.
Previous studies related the jet shift in response to tropical heating to the development of Rossby wave trains (e.g. Palmer andMansfield 1984, Ciasto et al 2016). Indications of Rossby waves originating from the tropics are seen in particular for the WP and EP cloud-radiative changes, and for these are consistent with the jet responses over the North Atlantic ocean (cf figures S4 and S5 for maps of stationary eddy stream function and meridional wind responses). However, even though all tropical regions show the zonal wind increase over Europe, they exhibit different responses of the stationary eddy stream function over Europe ( figure S4). Thus, we find no obvious link between the robust cloud-induced jet exit strengthening and Rossby wave trains originating from the tropics.
Our results highlight the importance of cloudradiative changes, especially those in the tropical upper troposphere, for the midlatitude circulation response to climate change. While cloud-radiative changes support the robust circulation response in some regions, they also contribute to uncertainties in the circulation response in other regions. Future studies should investigate the cloud-radiative impact in coupled climate models, and decompose the cloudradiative impact into the atmospheric and surface pathways.

Data availability statement
The data that support the findings of this study are openly available. The analysis scripts and the run scripts for the ICON simulations are provided in the Gitlab repository https://gitlab.phaidra.org/ albernn21/Albern-etal-clouds-jet-ERL2021 hosted by University of Vienna. Monthly-mean output from the ICON, MPI-ESM and IPSL-CM5A simulations that apply the cloud-locking and cloud-and water vapor-locking methods is published at KITopen with doi 10.5445/IR/1000134626. The KITopen data set also includes a copy of the analysis scripts and run scripts with git commit 9d03f05f7be7f785220c8f662fa64f3dd71a 52ec.