Future Climate Under CMIP6 Solar Activity Scenarios

Predictions of solar activity in the future are difficult to make due to the chaotic state of solar dynamo and the high nonlinearity of physical processes on the Sun. Therefore, the Climate Intercomparison Project Phase 6 (CMIP6) used a statistical approach and recommended two different solar forcing scenarios for the simulations. The reference scenario was developed as the standard forcing, whereas the alternative forcing has lower solar activity (EXT CMIP6). In this study, we use both forcings in a set of experiments to explore the importance of the alternative CMIP6 solar forcing for future climate and ozone layer variability. In general, the difference in solar forcing scenarios is small, and thus most changes at the surface and at high altitudes are not significant. In addition, only the active phases of the Sun, which have the largest difference in amplitude of the forcing, are investigated. In this case, some statistically significant patterns emerge, mostly in the stratosphere, but still, the magnitude of the changes is not very large and a noticeable surface climate response to these changes is not expected and also not found. Our results indicate that low amplitude solar forcings such as the EXT CMIP6 or similar are not worthwhile considering during the next CMIP type of activities. The proposed solar irradiance decline does not represent any danger to the ozone layer.

to June 2008. This recommendation followed Lean and Rind (2009) and was mainly used because solar projections were very uncertain at that time. Although the solar dynamo models have made progress in recent years (Charbonneau, 2014) they are still not precise enough to make satisfactory predictions of future solar activity. Thus, for developing future scenarios to be used in Climate Model Intercomparison Project Phase 6 (CMIP6), several, mainly statistical, approaches are used.
For CMIP6, the reference forcing (REF) is based on forecast models of the solar modulation potential (see Matthes et al. (2017) for details). The models used are an analogue forecast model, an autoregressive model, and a harmonic model with different training length and parameter settings to account for uncertainties. A modulation potential record from Steinhilber et al. (2012) was used as a training set and the final forcing is the weighted mean of the three different statistical methods. However, several statistical studies (Inceoglu et al., 2016;Steinhilber & Beer, 2013) suggested that grand maxima, as the Sun just exited recently, are generally followed by grand minima. Thus, Matthes et al. (2017) proposed an alternative scenario for sensitivity studies. The low activity scenario (EXT), also called Maunder-type scenario, is calculated using the lowest 5th-percentile of the solar modulation potential predictions. There is still a debate in the literature about what a low TSI might be like in the future and during grand minima in principle (Yeo et al., 2020). Egorova et al. (2018), Schmutz (2021), and Shapiro et al. (2011) argue that solar irradiance could be much lower and have more significant changes, but these estimates were not considered in the current CMIP recommendations. Mokhov et al. (2008) simulated climate change during the 21st century using several sets of anthropogenic and natural forcing. They showed that the surface climate response to solar irradiance forcing of up to 2 Wm −2 is rather small. Meehl et al. (2013) investigated the possible compensation of greenhouse warming by solar activity decline using the CESM1/WACCM model. They applied a TSI drop of about 3.9 Wm −2 from 2024 to 2065 and obtained up to 0.3 K cooling in 2070 after the end of the grand minimum of the solar activity. A continuation of this study with the same model but a less pronounced solar forcing decrease revealed some effects over the northern high latitudes during the cold season (Chiodo et al., 2016). Anet et al. (2013) and Arsenovic et al. (2018) exploited the SOCOL3-MPIOM model driven by a substantial decrease of the TSI (up to 6.5 Wm −2 ) according to Shapiro et al. (2011) and showed possible compensation of the greenhouse warming by up to 0.5 K and substantial delay with the future ozone layer recovery. A weaker solar forcing decrease (1.75 Wm −2 ) was used by Ineson et al. (2015) for the simulation of the future climate using the HadGEM2-CC model. They concluded that such a reduction can cool global climate only by 0.1 K during 2050-2100. There were no attempts to simulate the climate implications of the solar irradiance from the CMIP6 EXT scenario, though. Spiegl and Langematz (2020) forced the EMAC CCM model with a weak (1.3 Wm −2 ) and a strong (5.8 Wm −2 ) decrease of TSI and compared their results to an RCP 6.0 simulation. The global mean temperature at the end of the century is −0.1 K for the weak case and −0.45 K for the strong case.
The study presented here focuses on the impact of the two solar irradiance forcings for the future climate which were proposed for CMIP6. On the one hand, greenhouse gasses, mainly emitted close to the surface, influence the climate through longwave absorption and re-emission. On the other hand, the solar forcing "at the top" also modifies the climate as it influences ozone production in the stratosphere or irradiance on the Earth surface. In this study, we investigate the impact that the Sun might have on the climate over the 21st century.
Here, we use the Earth System Model (ESM) SOCOLv4 forced with the "middle-of-the-road" SSP2-4.5 scenario and the above-mentioned different TSIs and their respective spectral distribution as well as intensities of energetic particle fluxes. We briefly describe the model and methodology used, as well as the applied solar forcing, in more detail in the next Section. In Section 3, the results are outlined, starting from the temperature changes at the end of the century in Section 3.1, followed by Section 3.2, where the changes during the active phases are investigated in more detail. We conclude this study with a discussion and summary.

Model Description
For the simulations, we use the SOCOLv4 (SOlar Climate Ozone Links, version 4) ESM (hereinafter SOCOLv4). This model expands the Max Planck Institute for Meteorology (MPI-M) ESM version 1.2 (MPI-ESM1.2) (Mauritsen et al., 2019) with chemistry in the atmosphere. The MPI-ESM1.2 consists of a dynamical core, which is the general circulation model MA-ECHAM6, the interactive ocean model MPIOM, the ocean biogeochemistry model for carbon cycle HAMOCC, and the land surface model JSBACH. The MPI-ESM1.2 is coupled to the chemical module MEZON (Egorova et al., 2003;Rozanov et al., 1999) and the size-resolving (40 bins) sulfate aerosol microphysics module AER (Feinberg et al., 2019;Sheng et al., 2015;Weisenstein et al., 1997). The coupling occurs through the exchange of greenhouse gas concentrations, sulfate aerosol properties, and three-dimensional meteorological data such as relative humidity, clouds, and temperature. The SOCOLv4 is formulated on a Gaussian grid with a horizontal resolution of T63 spectral truncation (∼1.9° × 1.9°) as well as 47 vertical levels up to 0.01 hPa in a hybrid sigma-pressure coordinate system. The MEZON includes 99 chemical compounds, determined by 216 gas-phases, 72 photolysis, and 16 heterogeneous reactions. To compute dynamic processes, the SOCOLv4 uses a 15-min time step and a 2-hr time step to perform chemistry and radiation simulations. A lookup-table approach (Rozanov et al., 1999) is used to calculate photolysis rates, including the solar irradiance variability effect. The actual solar forcing in the model is divided into 14 wavelength bands used in the solar radiation module (Sukhodolov et al., 2014) and 78 bands for the photolysis and extra solar heating calculations (Sukhodolov et al., 2016). Further details of the model and an in-depth validation are described in Sukhodolov et al. (2021).

The Experiment Design and Solar Forcing Description
The model is run using historical boundary conditions until 2015 and the "middle-of-the-road" SSP2-4.5 scenario afterward (see Riahi et al. (2017) for a description of the SSP (Shared Socio-economic Pathway) scenarios). The solar forcing is branched into two different scenarios starting in 2020 and for each forcing we performed ensemble runs with three ensemble members each. Two additional shorter ensemble runs starting in 2050 were performed for the two solar forcing scenarios to better assess the variability of the climate response during the second half of the 21st century. For the analysis, all five ensemble members are used. The solar radiation scenarios used in our study are the REF and EXT described in Matthes et al. (2017). The corresponding TSI evolution is shown in Figure 1. Solar activity-related forcings such as the ionization by energetic particle precipitation (EPP) and NO x influx from the thermosphere also follow the recommendation by Matthes et al. (2017). No explicit eruptive volcanic forcing is used, however, the SSP scenarios include a background volcanic degassing forcing, as well as other surface emissions of sulfur compounds.
In the following, we analyze the result for two samples: the whole 20-year period between 2080 and 2099 and only periods where the Sun was in a high activity state. During the high activity phase, the difference between the REF and EXT forcings is the largest. These periods were defined as 5-to 6-year wide windows around the year where the REF solar cycle reached its maximum value. The following periods are selected: 2066-2071, 2079-2083, and 2090-2094 (gray bands in Figure 1). The mean differences in TSI during these periods are 0.46, 0.43, and 0.52 Wm 2 . The TSI differences between the two scenarios are remarkably similar to the amplitude of the REF scenario itself. Thus, the results from the highly active periods could also be interpreted as a future change between active and quiet periods during an 11-year solar cycle.
The significance is computed as in the IPCC AR6 report (see Gutiérrez et al., 2021, Cross-Chapter Box Atlas.1 (Approach C)). The variability threshold is computed as the standard deviation of the reference run (in our case REF) multiplied by the square root of two, a constant factor and divided by the square root of the period length. The square root of 2 accounts for the Gaussian propagation error as the variability of the two means could be different. The constant factor is in our case 1.645 and corresponds to a 90% confidence level. Thus, if the signal (in our case from EXT) exceeds 90% of the variability taking into account the propagation error and period length the signal is significant.

Changes at the End of the 21st Century
The difference between the two forcings in the global mean values, such as temperature or precipitation, is marginal. For instance, the annual mean temperature difference during the period 2080-2099, a typical IPCC report period, is 0.05 K (see Figure 2). Compared to the increase in warming between 2020 and the end of the century, which is around 2 K, the impact of the CMIP6-suggested solar forcing scenario is negligible and not significant. Similarly, the precipitation does not show any notable change until 2099 (not shown).
However, regionally there are some significant, although rare, temperature changes on the Earth's surface (see Figure 3). Most of the significant changes are present only for a specific season or month in a certain region as for example, the negative anomalies in summer and autumn in the Ross Sea. Previous studies which show

of 15
substantial changes at the Earth's surface usually apply much stronger forcing (e.g., Spiegl & Langematz, 2020). As described in the introduction, the forcing difference at the surface is on the order of a few 1/10th of a Wm −2 , which is small compared to other forcings. In order for small changes in TSI to have a significant effect on the surface temperature, there must be a mechanism that amplifies the response. Two such mechanisms are proposed in the literature, for example, the visible-driven "bottom-up" (i.e., Meehl et al., 2008;Meehl et al., 2009) and the UV-driven "top-down" (i.e., Kodera and Kuroda (2002)) mechanisms (see Maycock and Misios (2015) for an overview).
The "bottom-up" mechanism initiates with a temperature change at the surface, which will subsequently be modified through cloud and precipitation interactions (e.g., Meehl et al., 2008Meehl et al., , 2009). In our case, the forcing is too small to produce a significant impact on surface climate, related to the "bottom-up" mechanism. The only regions with consistently significant temperature changes are the Ross and Amundsen Sea close to Antarctica. These changes are probably indirectly induced by sea-ice cover changes.
The "top-down" mechanism is a sequence of changes induced by stratospheric warming in the tropics due to enhanced ozone production at the maximum of solar activity (see Figures A1 and A2 in Appendix A). The warming causes changes in the zonal mean temperature field and subsequently in the strength and location of the polar vortex due to the impact on the planetary wave propagation (i.e., see Gray et al. (2010) for an overview). Indeed, the zonal mean temperature changes show a large significant stratospheric cooling which travels from one polar region to the opposite polar region during their respective summer seasons (see Figure A2). This is expected as there is less incoming UV, less ozone production, and hence less warming in the sunlit stratosphere. Interestingly, these significant temperature anomalies are not changing the zonal wind and thus the vortex significantly ( Figure A3). Even though the changes in zonal wind seem quite large, the signal is still smaller than the variability. In autumn and spring, there are only a few patches spread in the stratosphere which show a statistically significant change in temperature.
As mentioned above, during the considered period only traces of the "top-down" nor the "bottom-up" mechanisms are visible, which agrees with Chiodo et al. (2016), who also used a weak future solar forcing scenario. The typical averaging period from 2080 to 2099 ranges roughly over two solar cycles. As the solar irradiances during the quiet phases are similar in the REF and EXT scenarios, the most extensive changes are expected due to a larger difference in the amplitude of the solar cycle. The mean TSI change at the end of the century is about 0.35 Wm −2 . This value is very small and lies within possible inter-monthly variability. Thus, in the second step, we investigate only the last three solar active phases of the 21st century. The results are described in the next section.

Changes During the Active Periods of Solar Irradiance
Narrowing the analysis only to the solar active phases does reveal some statistically significant features. The zonal mean ozone field shown in Figure 4 has a typical pattern of changes in the response to UV radiation decline (e.g., Arsenovic et al., 2018;Maycock et al., 2018). In the mesosphere, the reduced UV radiation produces less HO x from H 2 O photolysis suppressing the HO x -catalytic cycle of ozone destruction. In the stratosphere, there is less production of ozone due to abated O 2 -photolysis by reduced UV radiation below 242 nm, leading to 1%-3% lower ozone concentrations under the EXT solar forcing scenario. The decrease of ozone is present in every month and spans over the whole stratosphere. The largest significant patch in the mid-stratosphere is located in the southern high latitudes during the southern summer months and displaces to the northern high latitudes during the northern hemispheric summer. Additionally, there is a quasi-stable region in the troposphere south of 50°S with increased ozone concentrations. Calisto et al. (2011) reported a similar finding in the unpolluted Southern Hemisphere, mainly through the interplay of galactic cosmic rays (GCR), NO x , and ozone called the NO x -limited regime, that is, the ozone production becomes more sensitive to changes in NO x . In the Northern Hemisphere, the GCR-induced ozone changes in the troposphere are not present due to the atmosphere being more polluted with NO x . In addition, ozone in this region is strongly affected by the transport of ozone-rich air from the stratosphere (e.g., Revell et al., 2015). However, it is difficult to attribute this increased ozone signal to only one mechanism. Most likely it is caused by the interplay of all the above-mentioned factors.
Nevertheless, these variations in the ozone and reduction in irradiance available to be absorbed cause various changes in the temperature and wind fields ( Figure A3). The zonal mean temperature changes during the active phase show large significant responses in the mid-stratosphere (see Figure 5). Likewise, with the ozone changes described above, the significant changes in temperature follow the seasonal variation of the solar zenith angle and are therefore statistically significant only in the sunlit regions of the stratosphere (the summer hemisphere), where the dynamical variability is also much smaller than in the winter hemisphere. This pattern is also similar to the pattern observed during the period 2080-2099. Most of the statistically robust changes show a temperature decrease with a few exceptions. The positive ozone changes in the mesosphere due to the slow-down of the HO x -cycle (see Figure 4) do not strongly impact the mesospheric temperature signal, because ozone absorbs very little above 70 km. The upper mesosphere is dominated by the absorption of molecular oxygen in the Lyman-α line and the Schuman-Runge bands, which also get reduced due to less irradiance. However, the temperature changes due to dynamical variability are much larger in that region and therefore the solar signal is indistinguishable. This is consistent with results from high-top models which include O 2 absorption (e.g., Gruzdev et al., 2009).
In the northern latitudes, there is an effect that might be related to the "top-down" mechanism described by Kodera and Kuroda (2002). In addition, there is a non-significant warming close to the surface in the Arctic during February due to changes in solar forcing, similar to that reported by Drews et al. (2022). However, most temperature differences are non-significant. The zonal wind fields again show relatively large changes, but they all are also non-significant ( Figure A3). Similar to the climatology between 2080 and 2099, the variability is larger than the signal itself.
Even in the active phase, there are no large significant changes in temperature visible at the Earth's surface. There are only two main spots with substantial changes (see Figure 6a). One is located close to Antarctica and the other is in the North Pacific. During the austral winter and spring, the region with significant temperature change emerges near Antarctica in the Weddell and Ross Seas. A similar dipole structure in temperature response is often observed in the Southern Ocean sea ice and is correlated to the Southern Annular Mode (SAM) through pressure and advection changes (Lefebvre & Goosse, 2005). This is also the reason why the signal is more pronounced during the Southern Hemisphere winter-and springtime, as the sea ice in those regions melts to a large part during the summer.
A second significant temperature anomaly is located in the northern Pacific Ocean with positive values and the adjacent region of Alaska and the Rocky Mountains showing a negative anomaly, a typical picture of the Pacific Decadal Oscillation. Several studies (e.g., Chunhan et al., 2021;Guttu et al., 2021) have found a similar relationship between changes in TSI and the Pacific Decadal Oscillation through changes in temperature gradients induced by different UV values. Chiodo et al. (2016) found a similar change in temperature over the Pacific Ocean using idealized model experiments. They attributed the change mainly to the alteration of the visible part of the solar irradiance spectrum.
However, as shown in Figures 6b-6d, the variability is quite large among the three active periods considered. The anomaly in the North Pacific is visible in two periods (see Figures 6b and 6c). Also, the anomaly close to Antarctica is not always pronounced. Additionally, one period shows a stronger, statistically significant cooling over Siberia, which is not as pronounced in the mean over the three considered periods (Figure 6b). Gerber et al. (2014) and Zhang et al. (2018) report a link between Arctic warming and colder temperatures in Siberia. Depending on the baroclinicity over the Arctic and on the resolution of the stratosphere, a response over Siberia appears.
As seen above, each period by itself includes phenomena reported already in other studies. In the mean temperature, however, these signals are weakened and some of them become non-significant. Now, the question arises as to how these response patterns emerge. On the one hand, it could be that since the forcing difference is small, random long-term internal variability might emerge from other components, such as the ocean. On the other hand, it may happen that there is a locking or interplay of the variability for various components, and this is mostly driven by dynamics. Since the solar forcing difference is relatively weak, it is difficult to assess the cause of the variations, and this is beyond the scope of this study.

Discussion and Summary
We have performed two sets of transient ensemble simulations, using the ESM SOCOLv4 for the period 2020-2099, applying the SSP2-4.5 future socio-economic scenario and two different solar forcing scenarios as proposed in Matthes et al. (2017) for CMIP6. One forcing is proposed as the best assumption resulting from statistical methods using the past solar cycles, while the other forcing represents a long-term low activity scenario so-called grand minimum. The amplitude of the TSI variations under the future grand minimum scenario is, however, rather small, and basically mostly just the reduction of the 11-year cycle amplitude with only slight (<0.2 Wm −2 ) reductions of the minima periods. The TSI difference in the used forcings is largest during the active phase of the 11-year solar cycle. During the low activity phases, the TSI values of the two forcings are very similar. Thus, averaging over several solar cycles results in a very small forcing difference which is even smaller than the intra-monthly variations. As a consequence, the analysis of the mean climate state change at the end of the 21st century reveals very small and noisy responses. All major solar-related processes described in the literature such as the "bottom-up" (e.g., Meehl et al., 2008Meehl et al., , 2009Meehl et al., , 2013 and "top-down" mechanisms (Kodera & Kuroda, 2002;Mitchell et al., 2015) involving downward signal propagation in high latitudes are visible, but the signals are only marginally significant. Generally, the mechanisms described in the literature are found using very large solar forcing and are often not transient (e.g., Spiegl & Langematz, 2020).
The changes observed during the active phases are qualitatively very compatible to those observed over the period 2080-2099 but are more pronounced in amplitude and therefore, more statistically significant. This suggests that the mechanisms and patterns are generally the same for the long-term TSI reduction and the 11-year cycle effects. However, this influence is too weak to induce a substantial change in the near-surface climate.
In this study only simple statistical methodologies on ensemble averages are used. It might well be the case that a more thorough analysis using more sophisticated methods would provide different results. But the first order impact of a low TSI change remains small. Although this study is conducted with several ensemble runs to improve statistical interpretability it is only conducted with one state-of-the-art model. Including a variety of models would solidify the results. However, the main message would likely be similar.
An intriguing question arises if, for a certain TSI difference, the climate response depends on the average background state of the climate. For example, is the response to a difference of 1 Wm −2 in solar forcing different if the CO 2 concentration is closer to preindustrial times as compared to future CO 2 levels? A second interesting question is if the response to the 11-year solar cycle is different with a different background climate and which role the amplitude of the cycle plays in a warmer or colder climate.
These questions can be addressed in future studies. The current work demonstrates that the use of EXT CMIP6 or similar scenarios for future solar activity will not lead to new insights and should therefore not be recommended for any following CMIP activities. Given that all our future projections are always based on what we know from the past, instead of the low-forcing future runs, we would suggest having a closer look again on the past centennial changes, but using latest fully interactive models, climate proxy reconstructions, and a variety of available solar forcing estimates.

Acknowledgments
We (JS, TS, TE, AKD, and ER) thank the Swiss National Science Foundation for supporting this study through the Nr. 200020-182239 project POLE (Polar Ozone Layer Evolution). TE and ER acknowledge the support from the Karbacher Fonds, Graubünden, Switzerland. The work of ER and TS has been partly supported by the Ministry of Science and Higher Education of the Russian Federation under agreement no. 075-15-2021-583. Calculations were supported by a Grant from the Swiss National Supercomputing Center (CSCS) under projects S-901 (ID 154), S-1029 (ID 249), and S-903. Part of the model development was performed on the ETH Zürich cluster EULER. We thank Wenjuan Huo and two anonymous reviewers for their helpful comments and suggestions to improve the manuscript.