2.1 Simulations

The simulations used here were performed with the Community Earth System Model version 1 (CESM1), containing atmosphere, ocean, sea ice and land components. We briefly describe the atmospheric component of the model, WACCM (Mills et al., 2017). The WACCM resolution is 0.9^∘ (latitude) by 1.25^∘ (longitude), with 70 vertical levels up to a model top of 140 km. The model comprehensively represents stratospheric processes. The inclusion of interactive chemistry is one key improvement upon previous generations of models used to study sulfate geoengineering (e.g., Ferraro et al., 2015). Feedbacks from changing ozone concentrations have significant effects on the large-scale stratospheric circulation, especially on the quasi-biennial oscillation, in the geoengineering scenario of this study (Richter et al., 2017), and it has been shown that these feedbacks can considerably reduce the midlatitude jet shift response to increased CO₂ (Chiodo and Polvani, 2017, 2019). Interactive chemistry is also important for sulfate aerosol concentrations, which are prognostically determined by oxidation of SO₂ by OH to H₂SO₄ and subsequent microphysics within a modal aerosol scheme, the Modal Aerosol Module (MAM3; Mills et al., 2017). The simulated perturbation to radiative forcing following the eruption of Mount Pinatubo compares well to observed estimates, providing validation for the model's radiative effects of sulfate aerosol (Mills et al., 2017).

We analyze the GLENS simulations, which are fully described by Tilmes et al. (2018a). We consider three scenarios, all of which were performed under the following Representative Concentration Pathway 8.5 (RCP8.5) emissions scenario for greenhouse gases: (i) Base – 20 ensemble members performed between 2010–2030; (ii) RCP8.5 – three members of Base that were extended out to 2097; and (iii) GEO8.5 – 20 ensemble members with added stratospheric geoengineering performed between 2020–2099, which were branched off from the 20 Base members; this experiment is named “Geoengineering” in Tilmes et al. (2018a) and “GLENS” in a few other studies (e.g., Jiang et al., 2019; Simpson et al., 2019).

Geoengineering in the GEO8.5 runs is implemented as SO₂ injections at four locations (15^∘ N and 15^∘ S at 25 km and 30^∘ N and 30^∘ S at 22.8 km, at 180^∘ longitude). A primary aim of the experimental design was to limit the side effects of geoengineering. To this end, the GEO8.5 simulations aimed to maintain three annual mean surface temperature metrics at 2020 levels – the global mean temperature, the interhemispheric temperature gradient and the Equator-to-pole temperature gradient – using a feedback algorithm that annually adjusted SO₂ injection amounts (Kravitz et al., 2017). By the end of the GEO8.5 simulations, the total SO₂ injection rate is 52 Tg yr⁻¹, which is around five times the one-time injection used for WACCM simulations of the Mount Pinatubo eruption. The ensemble approach is a strength of these geoengineering simulations, which allows us to separate the forced geoengineering response (given by ensemble means) from the noise due to internal variability (determined from the spread across ensemble members; Deser et al., 2012).

To supplement GEO8.5, we analyze an additional set of simulations, namely the GEOHEAT_S runs described in Simpson et al. (2019), which we label GEOHEAT for simplicity. These aim to isolate the impact of stratospheric heating by the additional sulfate aerosols present in the GEO8.5 simulations from other factors, such as the longwave effects of increasing greenhouse gases and the shortwave effects of sulfate aerosols. The GEOHEAT runs were performed under Base conditions (average radiative forcing over 2010–2030) but with additional stratospheric heating rates derived from the last 20 years (2075–2095) of the GEO8.5 simulations. The GEOHEAT ensemble contains four members, each with a 20 year length¹. Essentially, the GEOHEAT response relative to Base will be compared to trends within GEO8.5, as described in the next subsection, in order to identify which GEO8.5 trends ultimately arise from stratospheric heating by the additional sulfate aerosols.

2.2 Trends and indices of dynamical variability

We compare trends within the RCP8.5, GEO8.5 and GEOHEAT simulations in order to apply regression methods which elucidate the coupling between stratospheric and tropospheric climate responses. Linear trends are appropriate since the climate responses under sulfate geoengineering analyzed here are approximately linear in time and in SO₂ injection rate, as noted in previous studies for temperature (global mean) and precipitation (Simpson et al., 2019; Tilmes et al., 2018a). There are some exceptions such as drying over the Mediterranean in winter, which mostly occurs later in the simulations (Simpson et al., 2019). Fields are seasonally averaged before trends are computed. We primarily analyze the NH wintertime (December–February; DJF) and, for comparison, the NH summertime (June–August; JJA). Trends within RCP8.5 and GEO8.5 are taken for the 75-year period between 2020 and 2095. For GEOHEAT, its difference with the Base climatology gives the 65-year response between average 2020 and 2085 conditions. An equivalent trend is calculated by dividing this response by 65 for comparison with RCP8.5 and GEO8.5 trends. In all cases, trends are shown per 30 years.

We compute the following two indices of NH variability:

1. The Northern Annular Mode (NAM) is used to quantify stratosphere–troposphere coupling. We adopt its common definition as the leading empirical orthogonal function (EOF) of geopotential height anomalies (Baldwin and Thompson, 2009; Gerber et al., 2010). The EOF is computed within Base, with concatenated geopotential height fields (20 years in 20 runs) in order to give the best available representation of model variability. The calculation is then performed independently at each pressure level as follows: the global mean is removed, the monthly mean climatology is removed, the seasonal average is taken over DJF and the EOF pattern is calculated over the region 20–90^∘ N using a square root of cos (latitude) weighting. We next perform a projection onto the leading EOF in each GEO8.5 simulation. Again, the global mean is removed, the Base monthly mean climatology is removed and the seasonal average is taken over DJF. The principal component (PC) time series is calculated by projecting these anomalies onto the leading EOF and standardizing with the corresponding Base PC standard deviation. We select the NAM at 50 hPa (NAM₅₀), which is a common lower stratospheric metric for investigating stratosphere–troposphere coupling.

2. The North Atlantic Oscillation (NAO) is used to diagnose atmospheric circulation changes over the North Atlantic. The NAO index is computed over the North Atlantic region of 20–80^∘ N, 90^∘ W–40^∘ E in the following two ways: (a) from the leading EOF of sea level pressure (SLP) anomalies to show its surface behavior and (b) from the leading EOF of zonal mean zonal wind anomalies at each pressure level over the North Atlantic region to diagnose stratosphere–troposphere coupling. Other details of the calculation are the same as for the NAM, with the exception that the global mean is not removed from the raw fields.

3.1 Surface air temperature and precipitation responses

The main goal of this paper is to explain the forced wintertime warming trends over Eurasia, Greenland and the North Atlantic which are simulated under sulfate geoengineering over the period between 2020–2095, as shown in Fig. 1a. This winter warming is found despite stabilization of the annual mean Equator-to-pole surface temperature gradient by the feedback control algorithm and contributes to a dampened seasonal cycle in temperature (Jiang et al., 2019). Continental warming is largely absent in summer (Fig. 1b) – this seasonality by itself suggests a dynamical influence of the stratosphere in winter. Furthermore, we show, in Fig. 1d, a distinct dipole response in Eurasian precipitation under geoengineering. There is statistically significant drying over the Mediterranean and southern Europe and wetting to the north over Scandinavia and above 50^∘ N in the North Atlantic (Fig. 1d; see also Simpson et al., 2019). Precipitation trends in summer are reversed over Eurasia relative to winter (Fig. 1e), which is again suggestive of a stratospheric dynamical influence in winter.

Figure 1Trends in near-surface air temperature and precipitation. Ensemble mean linear trends in (a, b, c) temperature (in degrees Celsius per 30 years) and (d, e, f) precipitation (millimeters per day per 30 years) in different seasons and experiments. (a, d) GEO8.5 for DJF (December, January and February), (b, e) GEO8.5 for JJA (June, July and August) and (c, f) RCP8.5 for DJF. Stippling indicates a lack of statistical significance at the 95 % confidence level under a one sample, two-sided Student's t test using the standard deviation across ensemble members.

To place these geoengineering responses into the broader context of other anthropogenic forcings, we emphasize that the magnitude of the winter warming in the GEO8.5 simulations amounts to approximately one-third of the forced warming under RCP8.5 (Fig. 1c). This makes clear that a feedback control algorithm that only maintains large-scale, zonal mean and annual mean temperatures does not successfully alleviate seasonal and local changes in surface temperature, as shown in previous studies for the GEO8.5 simulations (Jiang et al., 2019; Kravitz et al., 2017; Tilmes et al., 2018a). As for precipitation, trends under geoengineering are largely a cancellation, or slight reversal, of wetting trends in winter over the northern latitudes under rising greenhouse gases in RCP8.5 (Fig. 1f).

It is also of interest to contrast the winter warming under the geoengineering scenario to the potential response of midlatitude surface temperatures to large, low latitude volcanic eruptions. Consider, first, the signal-to-noise ratio by the end of the century in GEO8.5 (shown in Fig. 2a). Here, the signal is defined as the ensemble mean difference between GEO8.5 (2075–2095 average) and Base (2010–2030 average), and the noise is defined as interannual variability (computed as the standard deviation of DJF annual averages in Base across 20 years in 20 members). We find that the signal-to-noise ratio lies just around or below 1 in northern Eurasia, which means the signal is about 1σ of interannual variability (Fig. 2a). While a 1σ anomaly would be small in the context of observing the response in the first or second years after a volcanic eruption, here, the statistical significance of the forced response (ensemble mean relative to standard error in mean) is robust. We demonstrate this in Fig. 2b by showing the year in which ensemble mean trends beginning in 2020 become statistically significant (and remain significant) at the 95 % confidence level, i.e., with $X/\mathit{\{}\mathit{\sigma }/(\surd (N-\mathrm{1})\left)\mathit{\right\}}\mathit{>}\mathrm{2}$, where X is the ensemble mean trend, σ is the spread across ensemble members, and N is the number of ensemble members (20) (following Deser et al., 2012). We find that trends become significant around mid-century over Eurasia (Fig. 2b). The robustness of the forced response in the GEO8.5 simulations is also illustrated by the strong agreement across all 20 ensemble members for 2020–2095 trends, as shown in the Supplement Fig. S1 (the same is true for precipitation; see Fig. S2). Of course, for an individual realization, like we would observe in the real world, it would take much longer than the timescales shown in Fig. 2b to detect a significant trend. The robustness of the forced response to sulfate aerosol injections stands in stark contrast to the lack of a simulated forced response in the winter following the eruptions of Pinatubo and Krakatau (Polvani et al., 2019; Polvani and Camargo, 2020), and is a result of the sustained² and continuously increasing sulfate forcing in this scenario of geoengineering.

Figure 2Significance of the DJF near-surface air temperature response under geoengineering. (a) The signal-to-noise ratio of the end-of-century ensemble mean response in GEO8.5 (2075–2095) relative to Base (2010–2030). (b) The endpoint year in which the GEO8.5 ensemble mean trends beginning in 2020 become statistically significant from zero, and remain significant, at the 95 % confidence level.

3.2 Stratosphere–troposphere dynamical coupling

We here argue for the existence of a dynamical stratosphere–troposphere coupling pathway under geoengineering in the NH winter. We begin by showing ensemble mean DJF average trends in zonal mean temperature and zonal wind for GEO8.5, GEOHEAT and RCP8.5 in Fig. 3. Zonal wind trends in individual members are shown in Fig. S3.

Figure 3Trends in DJF zonal mean temperature and zonal winds. Shown are ensemble mean linear trends in temperature (a–c; shading; degrees Celsius per 30 years) and zonal wind (d–f; shading; meters per second per 30 years). Responses are shown for (a, d) GEO8.5, (b, e) GEOHEAT and (c, f) RCP8.5. Contours show the Base climatology (2010–2030). Stippling indicates trends that are not statistically significant at the 95 % confidence level under a one sample, two-sided Student's t test using the standard deviation across ensemble members. ΔT₅₀ values for each experiment give the ensemble mean and tropical average temperature trend (30^∘ S–30^∘ N) minus the NH polar cap average trend (60–90^∘ N) at 50 hPa.

Download

Figure 3a shows that the heating caused by additional sulfate aerosols in GEO8.5 peaks at just over 4 ^∘C per 30 years around 50 hPa, while greenhouse-gas-induced cooling trends occur at the same pressures over the North Pole. The ensemble mean tropical average (30^∘ S–30^∘ N) temperature trend minus the NH polar cap average (60–90^∘ N) trend at 50 hPa is 5.2 ^∘C per 30 years in this experiment. Consistent with a thermal wind balance response to this change in temperature gradient, there is a forced strengthening of the NH wintertime polar vortex (up to 5 m s⁻¹ per 30 years around 10 hPa) (Fig. 3d). Richter et al. (2018) found the same result for the first GEO8.5 ensemble member, and we here confirm the robustness of the forced zonal wind response across 20 ensemble members, particularly below 10 hPa (Fig. S3). The GEOHEAT experiment confirms the link between the tropical lower stratospheric warming and strengthening of the polar vortex, as also discussed in Simpson et al. (2019). The tropical lower stratospheric warming is broadly similar to GEO8.5 but with no polar cooling, since greenhouse gas concentrations are kept at Base levels in GEOHEAT (Fig. 3b). The smaller change in the meridional temperature gradient (3.5 ^∘C per 30 years) explains the weaker strengthening of the NH polar vortex (Fig. 3e) in GEOHEAT compared to GEO8.5.

In contrast, under climate change alone (RCP8.5), the most notable zonal wind response is a strengthening of the upper flanks of the subtropical jets (Fig. 3f), which is a robustly simulated response to increased greenhouse gas radiative forcing and the resulting enhanced warming of the upper troposphere (Fig. 3c; Lorenz and DeWeaver, 2007; Shepherd and McLandress, 2011; Manzini et al., 2014). We find no statistically significant trend in the NH stratospheric polar vortex (Fig. 3f). The lack of a polar vortex response is consistent with large intermodel spreads found by previous studies, with model responses differing in sign even under large greenhouse gas forcing scenarios such as RCP8.5 and 4xCO2 (Ayarzagüena et al., 2018, 2020; Manzini et al., 2014; Simpson et al., 2018). With the caveat of the single model nature of our study, the stronger NH polar vortex under sulfate geoengineering is therefore a robust and key difference in the forced climate response compared to increasing greenhouse gases alone.

While Fig. 3d and e show a clear strengthening of the NH stratospheric polar vortex under geoengineering, the tropospheric response appears weaker. However, the zonal mean view masks zonal asymmetries in the troposphere. So, we now focus on the region of interest, the North Atlantic. Figure 4 shows ensemble mean trends in the NAO index, computed as the leading principal component of zonal wind as a function of height in the 20–80^∘ N, 90^∘ W–40^∘ E region (see Sect. 2.2). When focusing on the NAO, what becomes apparent is a downward extension of the forced dynamical signal, i.e., positive ensemble mean NAO trends from the stratosphere to the troposphere under geoengineering in GEO8.5 (Fig. 4a). The connection of this response to tropical lower stratospheric heating is supported, once again, by the similar response in GEOHEAT (Fig. 4b). That this stratosphere–troposphere response is forced by sulfate geoengineering is further underscored by the negligible or negative trends found under RCP8.5 (Fig. 4c).

Figure 4Vertical profile of trends in DJF zonal wind characterizing the NAO. Shown are ensemble mean linear trends in (a) GEO8.5, (b) GEOHEAT and (c) RCP8.5, with error bars showing the 95 % confidence interval ($\pm \mathrm{2}\mathit{\sigma }/\surd N$, where N is the number of ensemble members). The 5th–95th percentile range is shown by horizontal black lines. In (b) GEOHEAT and (c) RCP8.5, the confidence interval and range are essentially the same.

Download

Since we are ultimately interested in the surface circulation response and the associated impacts, we show, in Fig. 5, ensemble mean trends in SLP and the surface NAO time series (based on SLP; see Sect. 2.2). Winter trends under geoengineering in GEO8.5 show a band of increasing SLP over Eurasia, with a reduction over the North Pole (Fig. 5a); the general pattern is found in almost every ensemble member (Fig. S4). These SLP trends project strongly onto the positive phase of the surface NAO index. We see that the index is generally positive and increases at a rate of 0.17 ± 0.07 per 30 years (Fig. 5d). Consistently, the GEOHEAT ensemble mean response is significantly above zero at 0.55 ± 0.27 (Fig. 5d, magenta bar; see Simpson et al., 2019, for further comparison of the zonal wind responses between GEOHEAT and GEO8.5 in the North Atlantic sector).

Figure 5Trends in SLP and the surface (SLP-based) NAO time series. (a, b, c) Ensemble mean linear trends in sea level pressure (hectopascals per 30 years). (d, e, f) Ensemble mean time series of the NAO. Responses are shown for (a, d) GEO8.5 (DJF), (b, e) GEO8.5 (JJA) and (c, f) RCP8.5 (DJF). The magenta point in (d) shows the ensemble mean response in GEOHEAT and its 95 % confidence interval. Stippling in (a, b, c) indicates lack of statistical significance at the 95 % confidence interval under a one sample, two-sided Student's t test using the standard deviation across ensemble members. The gray shading in (d, e, f) indicates the 95 % confidence interval ($\pm \mathrm{2}\mathit{\sigma }/\surd N$, where N is the number of ensemble members). The vertical scales for the NAO index are different in order to clearly illustrate the respective time series and confidence intervals.

Once more, we highlight the unique feature of the wintertime geoengineering response as compared to the response in RCP8.5, where no simulated trend is seen (0.02 ± 0.20 per 30 years) in the DJF NAO index (Fig. 5c, f). There is again also a distinct seasonality, with an opposite response under geoengineering in the summer, when the NAO index in fact decreases throughout the century with a trend of −0.28 ± 0.07 per 30 years (Fig. 5b, e). In the absence of the stratospheric polar vortex, the JJA response is likely related to ocean circulation changes in this model (Fasullo et al., 2018), although its robustness across other models remains to be determined.

3.3 Quantifying the impact of stratospheric dynamics on surface climate responses

Having shown substantial evidence connecting the stratospheric, tropospheric and surface wintertime circulation responses under geoengineering, we complete our analysis by determining what fraction of the surface temperature and precipitation responses, which we are ultimately interested in, this dynamical coupling can explain. First, within the GEO8.5 simulations, we regress, on a grid-cell-by-grid-cell basis, the DJF time series of each surface field against that of NAM₅₀ (following Thompson et al., 2000). The regression coefficient, multiplied by the trend in NAM₅₀, yields the part of the surface climate trend that is congruent with NAM₅₀. The residual is then the NAM₅₀-congruent trend subtracted from the total trend in the surface field. Second, to confirm which responses are ultimately caused by stratospheric heating from additional sulfate aerosols, we investigate the GEOHEAT responses. Under the proposed mechanism connecting tropical lower stratospheric heating, a strengthening of the NH polar vortex, downward dynamical coupling and circulation-driven climate changes, the GEOHEAT and NAM₅₀-congruent responses should be the same. Indeed, we will show that they are, to a reasonable extent.

Performing this analysis for the surface circulation response, we find, in Fig. 6a, that NAM₅₀-congruent trends in SLP indicate a more positive shift of the NAO than the full trend (compare to Fig. 5a), as seen in the deeper low over the Arctic and the high shifted towards the North Atlantic. Consistently, the NAM₅₀-congruent temperature and precipitation trends both depict dynamically driven responses to a positive NAO phase. There is warming over the Eurasian continent (Fig. 6d), which explains most of the full trend (Fig. 1a; upwards of 60 %). The dipole response of drying over southern Europe and wetting over northern Europe (Fig. 6g) is similar in pattern, but larger in magnitude, than the full trend (Fig. 1d). Furthermore, the SLP pattern, northern Eurasian warming and the dipole response in precipitation over western Europe in GEOHEAT (Fig. 6b, e, h) is very similar to NAM₅₀-congruent trends within GEO8.5 (Fig. 6a, d, g). We therefore conclude that downward dynamical stratosphere–troposphere coupling, which is ultimately driven by tropical lower stratospheric heating from sulfate aerosols, is the major driver of Eurasian winter warming, and associated changes in precipitation, in this particular scenario of sulfate geoengineering.

Figure 6Results from regressions of DJF surface climate responses against the stratospheric NAM in GEO8.5 and the responses in GEOHEAT. Ensemble mean linear trends in (a, b, c) sea level pressure (hectopascals per 30 years), (d, e, f) near-surface air temperature (degrees Celsius per 30 years) and (g, h, i) precipitation (millimeters per day per 30 years). Shown are trends in GEO8.5 that are congruent with NAM₅₀ (a, d, g), equivalent trends in GEOHEAT (b, e, h), and the difference between total GEO8.5 trends and NAM₅₀-congruent trends (c, f, i). Stippling indicates lack of statistical significance at the 95 % confidence interval under a one sample, two-sided Student's t test using the standard deviation across ensemble members.

Some large residuals from the NAM₅₀ regressions remain in the surface climate responses, which, therefore, cannot be explained by dynamical stratosphere–troposphere coupling. Residual SLP trends are large and oppositely signed to the NAM₅₀-congruent portion (compare Fig. 6a and c) and, thus, diminish the full response (Fig. 5a). There are residuals of around 1–2 ^∘C per 30 years in temperature trends over the Barents–Kara Sea, Greenland and the North Atlantic (Fig. 6f). The literature offer some explanation for these temperature residuals, as follows: annual mean forced warming around Greenland has been linked to changes in the hydrological cycle over the North Atlantic and an acceleration of the Atlantic Meridional Overturning Circulation (AMOC), although this might be a model-dependent feature (Fasullo et al., 2018). The warming over the Barents–Kara Sea is associated with sea ice losses in midwinter and spring (Jiang et al., 2019), but the feedbacks remain to be investigated. For precipitation, the residual is a drying over northwestern Europe and a wetting to the south (Fig. 6i), which also diminishes the full response (Fig. 1d) compared to the NAM₅₀-congruent response alone (Fig. 6g). The circulation residual (Fig. 6c) could be contributing to the precipitation residual, and there is also a possible role of an overall weakening of storm track activity under the combined influence of geoengineering and increasing greenhouse gases (Simpson et al., 2019).

Unlike for the North Atlantic and Eurasia, the stratospheric NAM cannot explain most of the climate responses to geoengineering over the Pacific Ocean and North America. The residual from the NAM₅₀ regression analysis shows reductions in North Pacific SLP (Fig. 6c), which combine with the opposing NAM₅₀-congruent portion (Fig. 6a) to cause a dipole pattern in the full GEO8.5 trend (Fig. 5a). There are also residual warming trends over North America (Fig. 6f) and a dipole response in precipitation over the North Pacific (Fig. 6i) that mostly explain the full trends (Fig. 1a, d). The deepening of the Aleutian low in the residual could signal long-term changes to the El Niño–Southern Oscillation (ENSO). As shown in Simpson et al. (2019), there are some enhancements in tropical East Pacific sea surface temperatures and reductions in the tropical West Pacific in GEO8.5, which are only partially explained by stratospheric heating. Changes in ENSO variability have also been reported in a solar dimming experiment (Malik et al., 2020) and following volcanic eruptions (Khodri et al., 2017). The Pacific climate response in GEO8.5 and under sulfate geoengineering in general merits future study.

We return to the main point that stratospheric dynamics are a key influence on the Eurasian surface climate in this scenario of geoengineering with a prominent seasonal wintertime signature. Without its influence, the residual from the NAM₅₀ regression suggests a negative NAO response under sulfate geoengineering in this particular model (Fig. 6c). Indeed, in the absence of the polar vortex in the NH summer, we have instead revealed a shift towards the opposite NAO phase (Fig. 5b, e).