2.1 The G6 experiments based on UKESM1

In this study, the G6sulfur and G6solar experiments in the sixth phase of GeoMIP are applied for analysing the AR characteristics under SAI. The G6sulfur experiment simulates future climate under the SSP5–8.5 scenario with stratospheric SO₂ injection aiming to reduce the global mean air temperature to the level under the SSP2–4.5 scenario (Kravitz et al., 2015; Jones et al., 2021, 2022). The SAI is continuous and applied between 10^∘ N and 10^∘ S along the Greenwich meridian at 18–20 km altitude. Such a tropical injection strategy has a minor impact on polar ozone (Tilmes et al., 2022), though it can overcool the injection sites relative to high latitudes and suppress global monsoon precipitation in contrast to the polar injection strategy (Sun et al., 2020). The injection rate is adjusted every 10 years to achieve a simulated decadal global mean temperature within ±0.2 K of that under the SSP2–4.5 medium forcing scenario. The G6solar experiment has the same goal but achieves it through the idealised gradual reduction in the solar constant (Kravitz et al., 2015; Visioni et al., 2021). Although such a solar dimming scenario is highly idealised compared to the SAI strategy prescribed by G6sulfur, it provides an important reference of a cooled future climate excluding the SAI-related aerosol–climate interaction; hence, its comparison with G6sulfur helps us understand the effect of SAI.

The G6 experiment outputs are from the UK Earth System Model (UKESM1; Sellar et al., 2019), a contributing model to the current Coupled Model Intercomparison Project Phase 6 (CMIP6; Eyring et al., 2016). The atmospheric component of UKESM1 is at a spatial resolution of $\mathrm{1.875}{}^{\circ }\times \mathrm{1.25}{}^{\circ }$ and 85 hybrid levels extending up to 85 km. The model applies the Even Newer Dynamics for General atmospheric modelling of the environment (ENDGame) dynamical core, a semi-implicit semi-Lagrangian scheme to solve the non-hydrostatic, fully compressible Navier–Stokes equations (Wood et al., 2014; Walters et al., 2019; Mulcahy et al., 2018). The atmospheric component is coupled to the NEMO (Nucleus for European Modelling of the Ocean) v3.6 ocean model with a spatial resolution of ∼1^∘ and 75 levels (Storkey et al., 2018). The land component applies the Joint UK Land Environment Simulator (JULES; Best et al., 2011). The atmospheric chemistry model in UKESM1 applies the United Kingdom Chemistry and Aerosols (UKCA) modelling framework and considers a stratospheric–tropospheric scheme with aerosol chemistry and online photolysis (Morgenstern et al., 2009; Mann et al., 2010; Archibald et al., 2020). The model also simulates the marine carbon cycle using a biogeochemical model described by Yool et al. (2013). For the experiments considered in this study, UKESM1 is run with three ensemble members (identifier index of r1i1p1f2, r4i1p1f2, and r8i1p1f2) corresponding to different realisation set-ups. The atmospheric river detection tool (ARDT) is first applied to simulations by each ensemble member. Then, the future changes in AR characteristics are analysed by calculating the ensemble mean differences between those identified in the future climate simulations (SSP5–8.5, SSP2–4.5, G6sulfur, and G6solar) for the period 2071–2100 and those identified during the baseline period 1981–2010 using the UKESM1 historical simulations for CMIP6. An evaluation of the performance of UKESM1 in simulating the climatology of the detected ARs and associated environmental fields and precipitation is given in the Supplement. Although some biases are noted, UKESM1 displays some consistency in the diagnosed AR properties with those identified in the historical climate reanalysis ERA5 (Hersbach et al., 2020) and observed precipitation (Hamada et al., 2011; Yatagai et al., 2014). For the ability to capture the modulation of ARs by the large-scale circulation, particularly for the East Asian jet stream (EAJS), comparisons of the correlation between AR frequency and the jet intensity are made between ERA5 and UKESM1 in the Supplement. It is found that UKESM1 reasonably captures the negative correlations between AR frequency to the north of 35^∘ N and the strength of westerly jet stream in terms of the East Asian jet stream index (EAJSI; Lu et al., 2011) compared to the ERA5 reanalysis dataset (Fig. S4a and b). Hence, the model can provide useful insight into the impact of SAI on ARs and its environmental drivers. Other information on the performance of UKESM1, e.g. biases in simulating the global distributions of temperature, precipitation, and aerosol optical depth, can be found in Sellar et al. (2019). In addition, at present, UKESM1 is the only model that provides outputs of pressure-level winds and specific humidity data every 6 h that satisfy the requirement of the used ARDT; hence, only the UKESM1 simulations are considered in this study, and other GCM ensembles conducted by the G6 experiments of GeoMIP (e.g. Jones et al., 2021, 2022; Visioni et al., 2021) are not used.

2.2 AR identification

The analyses of AR characteristics in gridded climate data rely on a variety of ARDTs. According to the current ARTMIP project (Shields et al., 2018; Rutz et al., 2019; O'Brien et al., 2022), the diversity of ARDTs is a major cause of the uncertainties in the diagnoses of AR metrics; hence, the use of ARDTs should be carefully selected and tuned for the specific research objectives. This study is based on the ARDT named Atmospheric River Identification Algorithm for the Asian monsoon region (ARIA-Asia). A detailed description of the algorithm procedures and their schematic can be found in Liang et al. (2022). In general, this ARDT comprises three steps, i.e. feature isolation, geometric analysis, and computation of diagnostic fields. As shown in Liang and Yong (2022), the first step calculates fields of vertically integrated water vapour transport (IVT) every 6 h and isolates the continuous regions that are likely AR plumes for each time step using fixed or spatially varying IVT thresholds (the latter is used in this study). The second step is to identify the axes of the detected features using the skeletonisation method for a binary image (Wick et al., 2013) and filter out the non-AR features (e.g. tropical cyclones and monsoonal flows) via the geometric criteria, i.e. the length of axis must be longer than 2000 km, and the equivalent width (the area–length ratio) must be less than 750 km. The third step is to calculate the diagnostic fields that quantify AR characteristics, including fields of AR frequency and AR-associated precipitation. This algorithm applies a filter of tropical moisture filaments, as seen in Liang et al. (2022) and Liang and Yong (2022); however, in contrast to these studies using an absolute (fixed) IVT threshold for the isolation of AR plumes, here we apply a relative thresholding method. First, following the methodology of ARDTs used in East Asia (e.g. Pan and Lu, 2019, 2020; Park et al., 2021; Kim et al., 2021), the algorithm uses spatially varying thresholds of IVT in terms of the 85th percentile of IVT averaged over a 5-month moving time window (Park et al., 2021) during the targeted 30-year periods. Second, as per Pan and Lu (2019, 2020), the Gaussian filter with a bandwidth of 6^∘ is applied to the IVT threshold fields. The lower limit of the thresholds (i.e. the lowest boundary value of the isolated AR plumes) is determined by the 80th percentile of IVT over the region from 40^∘ E to 120^∘ W and 20^∘ S to 60^∘ N. These help to discern the AR plumes from the large-scale background with relatively weak IVT and avoid the influence of high-frequency noise in IVT fields so that more coherent features can be obtained. In addition, the relative thresholding method scales the thresholds, respectively, for the historical and future climate simulations, which helps remove influences of the large-scale thermodynamic factors that are solely governed by the Clausius–Clapeyron relationship so that the dynamic changes in ARs due to different external climate forcings are focused.

3.1 Changes in AR-associated environments

Before analysing the impacts of SAI on the identified AR characteristics, such as the AR frequency and AR-related precipitation, we first assess the changes in AR-associated large-scale environmental fields and discuss their potential influences on ARs. Figure 1 shows the ensemble mean of absolute changes in the low-level, environmental fields averaged over the main AR season (May to September, MJJAS) for the future period (2071–2100) relative to the historical baseline period (1981–2010). The mean 850 hPa geopotential height during the historical baseline period shows the domination of the western Pacific subtropical high (WPSH), an important monsoon system that determines the spatial distribution of ARs (Pan and Lu, 2020; Park et al., 2021). Simulations under the SSP5–8.5 scenario project significant increases in geopotential height to the south of 25^∘ N (maximum change centred around 20^∘ N, 85^∘ E; coloured in Fig. 1a) accompanied with an enhancement of low-level convergence across 30–45^∘ N. Such changes are associated with an increase in the meridional land–sea contrast in terms of the surface temperatures (Fig. S5). These drive an intensification of the southwesterly monsoon across about 20^∘ N and lead to stronger stationary moisture transport from the moisture source region over the Bay of Bengal, which implies a more favourable environment for the upstream AR activities. Similar patterns are noted under SSP2–4.5 (Fig. 1b), while the magnitude of increases in the 850 hPa geopotential height is reduced by about one-third compared to SSP5–8.5. Both G6sulfur (Fig. 1c) and G6solar (Fig. 1d) show ameliorated changes with respect to SSP5–8.5. Although the SSP2–4.5, G6sulfur, and G6solar experiments present similar magnitudes of changes, there are noticeable differences that are investigated further.

Figure 1Absolute changes in MJJAS mean geopotential height (shaded; unit in geopotential metres or gpm) and wind (vectors) at 850 hPa for the future period of 2071–2100 relative to the historical baseline during 1981–2010. Black contours show the MJJAS mean 850 hPa geopotential height during the historical baseline period. Areas with a surface pressure below the 850 hPa level are shown in white.

Figure 2Differences in MJJAS mean geopotential height (shaded) and wind (vectors) at 850 hPa for the future period of 2071–2100 between G6sulfur and SSP5–8.5 (a) and SSP2–4.5 (b) and G6solar (c).

Figure 2 shows comparisons of the simulated future low-level circulations between G6sulfur and other experiments. Compared with SSP5–8.5 (Fig. 2a), the simulated SAI by G6sulfur leads to smaller increases in the geopotential height at 850 hPa over most of the study region, despite the North Pacific (near 40^∘ N, 165^∘ E), where an anticyclonic anomaly (in terms of the 850 hPa wind field) is observed, and this favours the northeastward moisture transport of ARs across 40–55^∘ N. Compared to SSP2–4.5, G6sulfur shows negative anomalies of the geopotential height over the continent (Fig. 2b). Such a change corresponds with a more pronounced land–sea thermal contrast, with a significant warming (p value < 0.05) of the surface temperature to the north of 30^∘ N and cooling to the south (Fig. 3b). These changes drive an intensification of the southwesterly monsoon flow and the related cyclonic shear across the upstream of the main AR-active region (25–30^∘ N, 105–120^∘ E). In addition, the low-level anticyclonic response over the North Pacific is corresponding with a significant warming of the sea surface temperature (SST) by up to 1.5 K. These changes may favour the AR-associated blocking at upper midlatitudes (Mullen, 1989; Pohl et al., 2021). Similar patterns of circulation differences are presented when comparing G6sulfur with G6solar (Figs. 2c and 3c). Therefore, although G6sulfur simulates ameliorated changes in the low-level circulation compared to SSP5–8.5, the low-level environments are more favourable for AR activity for G6sulfur compared to SSP2–4.5 and G6solar. Thus, although the simulated SAI strategy is successful at reducing many of the changes apparent under SSP5–8.5, some differences are evident between G6sulfur, G6solar, and SSP2–4.5 (which all have the same global mean temperature), which is due to the existence of the aerosol–climate feedback under SAI.

Figure 3As in Fig. 2 but for the MJJAS mean surface temperature. Stippling indicates changes that are statistically insignificant (p value > 0.05 based on Student's t test).

The meridional displacement of ARs is related to various high-tropospheric systems, including the westerly jet streams and the associated upper-level high pressure (Payne and Magnusdottir, 2015; Liang et al., 2022). Following the analyses of Kamae et al. (2017b) and Liang et al. (2022), Fig. 4 assesses changes in the AR-related upper-level environments, including the EAJS, in terms of the 200 hPa wind field and the upper-level thermal expansion quantified by 200–500 hPa geopotential thickness. For the warming scenario under SSP5–8.5 (Fig. 3a), significant increases (p value < 0.05) in the thermal expansion are observed over the study region, with a maximum increase of more than 230 gpm near northeastern China (130^∘ E, 50^∘ N) and to the south of the WPSH centre (170^∘ W, 20^∘ N). The increased thermal expansion to the north of 45^∘ N leads to a decreased meridional thickness gradient and consequently drives a significant weakening (by up to 5 m s⁻¹) of the EAJS across 30–45^∘ N. Similar changes in the thermal expansion (Kamae et al., 2014) and EAJS (Endo et al., 2018) related to the increase in the meridional land–sea thermal contrast have been found under the radiative forcing of the 4×CO₂ experiments, while opposite changes have been observed under only the forcing of the SST warming. The projected weakening of the EAJS favours a northward shift in ARs, according to the correlation analyses by Liang et al. (2022) and those discussed later (Fig. S4; discussed in Sect. 3.2). Smaller magnitudes of changes (by about 90 gpm) in the geopotential thickness are shown over most of the region under SSP2–4.5 (Fig. 4b), and this also results in a less apparent weakening of the EAJS. However, G6sulfur projects the most pronounced weakening of the EAJS (by up to 8 m s⁻¹) due to the significant decrease in the thickness gradient between 30–45^∘ N, which is associated with the reduced increases in the geopotential thickness to the south of 30^∘ N. The further weakening of the EAJS is also related to the presence of the increased maximum to the north (near 130^∘ E, 50^∘ N), though the magnitude is smaller (by about 70 gpm) than that in SSP5–8.5. In contrast, G6solar (Fig. 4d) projects similar patterns and magnitudes of the changes in the upper-level thermal expansion and winds compared to SSP2–4.5. To further present the effect of SAI on the AR-related upper-level environments, the absolute differences, of 200–500 hPa geopotential thickness and 200 hPa winds, between G6sulfur and other future climate simulations are displayed (Fig. 5). The comparisons confirm that SAI can lead to a further weakening of the EAJS in a warmer climate, even though it effectively ameliorates the increase in the upper-level thermal expansion of the high-emission scenario (shown by the negative red contours in Fig. 5a). Compared with SSP2–4.5 and G6solar, the injected sulfur dioxide simulated by G6sulfur leads to further decreases in the intensity of the EAJS, and these are driven by the reduction in the meridional thickness gradient across 30–45^∘ N (Fig. 5b and c).

Figure 4Absolute changes in MJJAS mean geopotential thickness between 200 and 500 hPa (red contours; unit is gpm) and 200 hPa wind speeds (shaded) for the future period of 2071–2100 relative to the historical baseline during 1981–2010. Black contours (unit in gpm) show the MJJAS mean 200–500 hPa geopotential thickness, and vectors show the MJJAS mean 200 hPa wind fields during the historical baseline period.

Figure 5Differences in MJJAS mean geopotential thickness between 200 and 500 hPa (red contours) and 200 hPa winds (shaded for velocity and vectors for direction) for the future period of 2071–2100 between G6sulfur and SSP5–8.5 (a) and SSP2–4.5 (b) and G6solar (c).

As the variation in AR frequency over East Asia is closely associated with the North Pacific storm track (Ryu et al., 2021), the absolute differences in the root mean square field of 2–6 d bandpass-filtered mean sea level pressure (MSLP) between G6sulfur and other future climate simulations are analysed as a proxy for eddy kinetic energy and storm tracks (Fig. 6). Such an analysis is based on the Lanczos bandpass filter, following Harvey et al. (2020). For the main AR-active season (MJJAS), all the future climate simulations capture a notable reduction in the midlatitude storm tracks (Fig. S6), with the weakening of the EAJS in a warmer climate, a result consistent with the study of Harvey et al. (2020). Compared with SSP5–8.5, the injected sulfur dioxide simulated by G6sulfur shows no apparent amelioration of the decreases in the storm tracks (Fig. 6a), particularly to the south of Japan. It also presents further decreases across most of the North Pacific compared to SSP2–4.5 (Fig. 6b) and G6solar (Fig. 6c), and these changes can be explained by the weakened baroclinicity under the decreased meridional gradient of surface temperature across 20–50^∘ N (Fig. 3). Thus, the changes in storm tracks under the simulated SAI may further inhibit the transient AR features over the downstream region.

Figure 6As in Fig. 5 but for the root mean square of 2–6 d bandpass-filtered MSLP (proxy for eddy kinetic energy and storm tracks; unit in hPa) during MJJAS. Stippling indicates changes that are statistically significant (p value < 0.05 based on Student's t test).

In summary, the UKESM1 simulations under the SSP5–8.5 high-emission scenario project a strengthening of the upstream monsoon flows and the low-level convergence that are favourable for AR activity in most of East Asia. In addition, a weakening of the EAJS driven by the increased upper-level thermal expansion at high latitudes is displayed. The experiments SSP2–4.5, G6sulfur, and G6solar show ameliorated changes in the low-level circulation associated with ARs; however, compared to SSP2–4.5 and G6solar, the simulated deployment of SAI in G6sulfur exacerbates the weakening of the upper-level westerly jet, which is linked to the cooling effect of the injected aerosol precursor, concentrated across the lower latitudes. This is evident through the comparisons between G6sulfur and G6solar in terms of the global distributions of aerosol optical depth determined at 550 nm and the surface air temperature according to the study of Jones et al. (2021). Moreover, the simulated SAI by G6sulfur shows a limited effect on reducing the general decrease in the storm track across the North Pacific, which creates a less favourable environment for the downstream AR activity.

Figure 7Absolute changes (shaded) in annual mean AR frequency (percent of time steps every 6 h) for the future period of 2071–2100 relative to the historical baseline during 1981–2010. Black contours show the annual mean AR frequency during the historical baseline period. Stippling indicates changes that are statistically significant at a confidence level of >95 % (p value < 0.05 based on Student's t test).

Figure 8Differences in annual mean AR frequency for the future period of 2071–2100 between G6sulfur and SSP5–8.5 (a) and SSP2–4.5 (b) and G6solar (c). Stippling indicates differences that are statistically significant at a confidence level of >95 % (p value < 0.05 based on Student's t test).

3.2 Changes in AR features

Now we analyse the future changes in AR features. Under the SSP5–8.5 scenario (Fig. 7a), the ensemble mean of UKESM1 projects a significant increase (p value < 0.05) in the AR frequency over most of regions from southern China to the Korean Peninsula. The greatest magnitude of increase (by above 0.3 %) is seen in southern China. Significant decreases in frequency are shown over the western Pacific, which is possibly associated with the domination of the stable warm high-pressure within WPSH (Fig. 1a) and the less active storm tracks (Fig. S6a). The significant increase in the AR frequency is consistent with the more favourable dynamical condition in the low troposphere, as displayed in Fig. 1a, including the intensification of the southwesterly monsoon flow and the midlatitude low-level convergence. Under SSP2–4.5, a similar pattern of the frequency change to SSP5–8.5 is projected but with smaller magnitudes (Fig. 7b). The experiments of G6sulfur (Fig. 7c) and G6solar (Fig. 7d) show generally ameliorated changes compared to SSP5–8.5. However, generally, any amelioration of AR frequency is the lowest under G6sulfur, and G6sulfur shows opposite changes (increase by up to 0.2 %) to the north of the main AR-active region, implying a pronounced northward shift in ARs under SAI. This change is linked to the further weakening of the EAJS (Fig. 5c) given the negative correlation between AR frequency and EAJSI to the north of 35^∘ N ,as presented by both the ERA5 reanalysis dataset (Fig. S4a) and UKESM1 (Fig. S4b); however, underestimation of such a correlation in UKESM1 is noted to the east of 135^∘ E (Fig. S4c), and this bias is partly related to the underestimated AR frequency over the downstream region (Fig. S1c). Similar negative correlations between high-level jets and the upper-midlatitude AR frequency have been found by Zhang and Villarini (2018) and Liang et al. (2022). The mechanism behind the enhanced upper-midlatitude ARs by the weakening of EAJS remains elusive, though one of the possible causes is the anomalous low-level convergence to the northwest of the jet core and divergence to the northeast, according to the four-quadrant strait jet model (Uccellini and Johnson, 1979). This consequently favours the AR-associated northeastward transport of warm moist air across the northern flank of the EAJS. It is also noted that the further decreases in the storm track activity under the simulated SAI do not explain the increases in upper-midlatitude ARs over land, as these systems are mainly controlled by the anticyclonic quasi-stationary circulation (Park et al., 2021).

Figure 8 further illustrates how SAI influences the distribution of the AR frequency in terms of differences between G6sulfur and other experiments. Figure 8a shows that SAI ameliorates some changes to the south of 40^∘ N under SSP5–8.5, while it induces significant increases (by up to 0.15 %) in the AR frequency over northeastern China. Compared to SSP2–4.5 and G6solar, G6sulfur demonstrates further frequency increases over most of northeastern China and northern Japan (Fig. 8b and c), which are partly linked to the more pronounced intensification of upstream monsoonal flow and low-level cyclonic shear (Fig. 2b and c).

In summary, the SAI strategy simulated by the G6sulfur experiments can partly mitigate the changes in AR features in East Asia, particularly for the increase in AR frequency over southern China and decreases at the lower latitudes. However, SAI could potentially induce increases in AR activity across northeastern China and Japan as it further weakens the EAJS intensity that is negatively correlated to the upper-midlatitude AR activity. The presented AR–SAI connections could possibly lead to changes in the mean and extreme precipitation of the region, and this will be examined in the following section.

Figure 9Absolute changes in the annual accumulation of AR-related precipitation (a–d) and the fractional contribution of ARs to annual total precipitation amount (e–h) for the future period (2071–2100) relative to the historical baseline (1981–2010). Stippling indicates changes that are statistically significant at a confidence level of >95 % (p value < 0.05 based on Student's t test).

3.3 Changes in AR-associated precipitation

Previous sections have discussed the impact of SAI on the distribution of AR frequency and the associated large-scale environments in East Asia. This section further examines the responses of AR-associated mean and extreme precipitation under the future changes in ARs. Here, precipitation at a given location is considered to be AR-associated if an AR axis is within 350 km. For the SSP5–8.5 scenario (2071–2100) relative to the historical baseline period (1981–2010), the ensemble mean of UKESM1 projects a significant increase (p value < 0.05) in the annual mean accumulation of AR-associated precipitation across the Korean Peninsula and most of China, with magnitudes of up to above 320 mm yr⁻¹ (Fig. 9a). Similar changes with smaller magnitudes are seen under the SSP2–4.5 scenario (Fig. 9b). Resembling the changes in AR frequency, both G6sulfur (Fig. 9c) and G6solar (Fig. 9d) show similar magnitudes of increases across 30^∘ N compared to SSP2–4.5. For the changes in the fractional contribution of ARs to annual total precipitation, ARs tend to contribute more precipitation across central and eastern China (by up to 9 %) and North Korea (up to 5 %) under SSP5–8.5 (Fig. 9e). Decreases in the fraction are seen over northeastern China due to the pronounced increases in non-AR precipitation, which is not the focus of this paper. SSP2–4.5 (Fig. 9f), G6sulfur (Fig. 9g), and G6solar (Fig. 9h) project less apparent increases in the fraction to the south of 30^∘ N compared to SSP5–8.5. However, significant increases in the fraction by up to 6 % are seen in northeastern China and North Korea under the simulated deployment of SAI in G6sulfur.

For the ensemble mean of the simulated heavy rain events associated with ARs (which is the number of days accumulated when the daily precipitation amount is greater than 40 mm d⁻¹ at a given location and an AR axis is within 350 km), the patterns of changes shown in Fig. 10a–d resemble those for precipitation accumulation (Fig. 9a–d). It is noted that G6sulfur (Fig. 10c) shows significant increases in AR-associated extreme precipitation in Japan with a magnitude similar to that under SSP5–8.5 (Fig. 10a). Also, G6sulfur generally projects a larger magnitude of increases in the fraction of heavy rain events over most of northeastern China (Fig. 10g) compared to other future climate simulations (Fig. 10e, f, and h). These changes imply pronounced side effects of SAI on hydrological extremes from ARs in the above-mentioned regions.

Figure 10As in Fig. 9 but for the occurrence of AR-related events (a–d) of heavy rain (daily precipitation > 40 mm) and the contribution of ARs to the annual total heavy rain events (e–h).

We now compare the ensemble mean of the AR-associated precipitation patterns between G6sulfur and other experiments. As expected, G6sulfur effectively ameliorates the increase in AR precipitation across 30^∘ N compared to SSP5–8.5 (Fig. 11a). Figure 11b shows the comparison between G6sulfur and SSP2–4.5, which indicates a higher amount of AR precipitation for most of the upper midlatitudes, particularly central–northern China (by up to 160 mm yr⁻¹). Likewise, the comparison between G6sulfur and G6solar (Fig. 11c) indicates greater AR precipitation (by up to 100 mm yr⁻¹) for the same regions. For the AR-related heavy rain events, although ameliorated increases are shown in central and eastern China when comparing G6sulfur with SSP5–8.5 (Fig. 11d), the simulated SAI strategy exhibits amplified increases in the events in central–northern China and the Korean Peninsula with respect to SSP2–4.5 (Fig. 11e) and G6solar (Fig. 11f), implying a considerable side effect of SAI on extreme precipitation.

Figure 11Differences in annual accumulation of AR-related precipitation (a–c) and heavy rain events (d–f) for the future period of 2071–2100 between G6sulfur and SSP5–8.5 (a, d) and SSP2–4.5 (b, e) and G6solar (c, f). Stippling indicates differences that are statistically significant at a confidence level > 95 % (p value < 0.05 based on Student's t test).