For each of the four participating models, we present two sets of simulations: an ensemble modeling the SSP2-4.5 global warming scenario without SAI, and an ensemble modeling SSP2-4.5 in which SAI is used to cool the planet in accordance with the protocols of the G6-1.5K-SAI experiment , described below. SSP2-4.5, part of the Shared Socioeconomic Pathway framework used in CMIP6, is a moderate-warming, “middle-of-the-road” scenario in which GHG emission trajectories do not markedly depart from historical trends. Of the primary SSP scenarios, SSP2-4.5 most closely matches emissions pledges in the near- to medium-term future . The simulations of SSP2-4.5 considered in this study begin in model year 2015, and we consider data from the beginning of model year 2015 through the end of model year 2084.

In the G6-1.5K-SAI experiment, SAI is used to offset the warming in the SSP2-4.5 scenario. This is implemented in each model by placing SO₂ into the gridboxes nearest 30° N and 30° S latitude at approximately 21 km altitude. Injection begins in model year 2035 and continues through the end of model year 2084. The amount of injected SO₂ is chosen each year to regulate global mean near-surface air temperature (GMSAT); for each model, a target GMSAT is defined, equivalent to the SSP2-4.5 ensemble mean GMSAT averaged over the 2020–2039 period (serving as a proxy for 1.5 °C above preindustrial, as in ), and in each G6-1.5K-SAI simulation, injection rates are chosen each year to offset warming above this temperature target, with the total SO₂ divided evenly between the 30° N and 30° S injection sites. The necessary injection rates are computed separately for each ensemble member. To date, all participating models use a feedback control algorithm to compute the necessary injection rates , but this is done for convenience and is not strictly necessary. A more complete discussion of the major design decisions of G6-1.5K-SAI, and the reasoning behind them, can be found in .

Thus far, four ESMs have participated in the experiment: in alphabetical order, they are version 2 of the Community Earth System Model, CESM2(WACCM6) (henceforth “CESM”); version 3 of the Energy Exascale Earth System Model, E3SMv3 (henceforth “E3SM”); the Model for Interdisciplinary Research on Climate, MIROC-ES2H (henceforth “MIROC”); and version 1.1 of the UK Earth System Model, UKESM1.1 (henceforth “UKESM”). These four ESMs, their configurations, experimental ensemble sizes, and specifications for SO₂ injection are described in Table 1. Additional information about their aerosol representations is presented in Sect. 2.3.

All four participating models use modal aerosol representation, wherein aerosol populations are described by multiple “modes” with prescribed log-normal size distributions. The configurations of CESM and E3SM used in this study both use updated versions of the Modal Aerosol Module version 3 (MAM3), first presented by ; MAM3 represents sulfate aerosols using Aitken, accumulation, and coarse modes. CESM uses MAM4 , which uses the same three modes for sulfate as MAM3, with some subsequent modifications to size distributions described by ; the fourth mode (nucleation) is not used to represent sulfate. E3SM uses MAM5-PSA (“Prognostic stratospheric aerosol,” ), which uses the same three modes for sulfate as MAM4 (with some size modifications from ) while adding an additional coarse mode specifically for stratospheric sulfate. MIROC represents aerosol chemistry using the CHASER atmospheric chemistry model, and represents aerosol transport using the SPRINTARS aerosol transport model . This configuration represents sulfate aerosols with nucleation, Aitken, and accumulation modes. Importantly, the aerosols are not fully coupled with the radiation such that during radiation calculations, all aerosols are treated as a single mode with a prescribed size distribution. UKESM uses the five-mode version of the two-moment GLOMAP aerosol scheme (Global Model of Aerosol Processes; ), which represents sulfate using nucleation, soluble Aitken, accumulation, and coarse modes (the fifth mode, insoluble Aitken, does not include sulfate). Size parameters for each mode for each participating model are included in Table 2 below.

In Fig. 1, we present the evolution of global mean near-surface air temperature (GMSAT) for participating models. The models have a range of baseline climate states (and thus temperature targets) and rates of warming; the temperature targets cover a spread of approximately 0.5 °C, and on average, the models predict 0.30 ± 0.07 °C of warming per decade in the SSP2-4.5 scenario (where ± indicates one standard deviation across the four model ensemble means; this measure of uncertainty is used henceforth). In all participating models, SAI cools the planet (1.43 ± 0.35 °C), and the chosen injection rates reduce GMSAT to approximately maintain the temperature target. In any given year, there is a non-zero amount of error, or difference between each model's temperature target and the actual GMSAT; the difference across models in which the injections are able to maintain the temperature targets is a function not only of natural variability and aerosol behavior in each model, but also of the tuning of the controller used to choose injection rates. Diagnostics of this error are presented in Sect. S1 of the Supplementary Material. For the purposes of this study, any analysis which considers the amount of cooling (e.g., cooling per unit AOD in Fig. 2e, maps of cooling in Figs. 5–6) will consider the amount of actual cooling effected by the intervention in each model (i.e., SSP2-4.5 temperature minus G6-1.5K-SAI temperature) and not the intended cooling (i.e., SSP2-4.5 temperature minus the temperature target).

Figure 2 denotes the magnitude of the intervention required to offset warming above the GMSAT target in each model. Averaged over the last two decades of the experimental period (2065–2084), the models require 12.53 ± 2.82 Tg SO₂ injection per year, producing 0.15 ± 0.06 stratospheric 550 nm aerosol optical depth (henceforth “AOD”), to cool the planet by the desired margin. The amount of SO₂ needed to cool the planet in each model is a function of both the amount of warming to be offset and how effectively aerosols cools the surface; the latter is governed by the models' respective aerosol microphysics, chemistry and transport representations (which determine the timescale for SO₂ oxidation, aerosol transport, particle growth, absorption and scattering of visible and infrared radiation, and particle lifetime) and internal climate feedbacks.

While there is considerable uncertainty in the required total injection rates and AOD across models (Fig. 2a–b), the models strongly agree on the overall ratio of cooling per unit injection (Fig. 2c; 0.11 ± 0.01 °C (Tg SO₂ yr⁻¹)⁻¹). This is within the commonly-cited 8–16 Tg SO₂ yr⁻¹ °C⁻¹ range for SAI ; however, given the disagreement in AOD per unit injection (2d) and cooling per unit AOD (2e), and that the models use similar aerosol representation and are trained on the same volcanic eruptions, the agreement seen in 2c may not imply accuracy. Of the four participating models, CESM and UKESM show higher AOD per unit injection (2d) but lower cooling per unit AOD (2e), while E3SM and MIROC show the reverse. As a whole, Fig. 2c suggests higher cooling per unit injection and tighter model agreement for G6-1.5K-SAI than for G6sulfur; however, this is mainly because of the lower cooling per unit injection in the MPI models in G6sulfur, which disagree with the other three. The four G6-1.5K-SAI models cool the planet at similar rates per unit injection to the three G6sulfur models with interactive SO₂ injection, but as these models also include CESM2 and an earlier version of UKESM which also used GLOMAP, this agreement again may not imply accuracy. We elaborate more on this in Sect. 4.

The three interactive-SO₂ G6sulfur models produced, as a whole, similar or more AOD per unit injection than G6-1.5K-SAI models (2d). The literature disagrees on whether subtropical injection would produce AOD more or less efficiently than equatorial injection; due to the poleward direction of the stratospheric Brewer-Dobson circulation, stratospheric aerosols tend to be carried towards the higher latiutdes, however, the lifetime of aerosols decreases as aerosols are injected away from the tropical pipe and into the subtropics. report more AOD for prescribed equatorial injection in a multi-model comparison, but report higher AOD for 30° N + 30° S injection than for equatorial when used to cool the planet in CESM2.

In a longer-term injection scenario, larger concentrations of aerosols (such as due to aerosols being confined to the tropical pipe, as is common for equatorial injections) lead to increased particle size due to coagulation, decreasing particle scattering efficacy and lifetime e.g.,, suggesting lower AOD per unit injection for equatorial injection as rates of injection increase. Earlier modeling studies e.g., have found that injection efficiency begins to fall off with increasing injection rates beginning at 10–20 Tg SO₂ yr⁻¹; however, for G6sulfur, the downward trend of AOD production per unit injection with time suggests such a nonlinearity even at lower injection rates in the case of equatorial injection, and this trend is absent in G6-1.5K-SAI models except CESM, suggesting such nonlinearities are smaller or absent in this regime of injection rates in these models for subtropical injection. Cooling per unit AOD (2e) follows a similar trend for the two experiments, with a larger model spread for G6-1.5K-SAI than G6sulfur.

Figure 3 plots zonal mean AOD for G6-1.5K-SAI and G6sulfur. In all G6-1.5K-SAI participating models, AOD distribution shares the same broad characteristics, consistent with the poleward direction of the stratospheric Brewer-Dobson Circulation and the transport barriers of the stratospheric polar vortices: AOD increases poleward of the injection sites, peaks near the boundary of the polar vortex, and then plateaus (Northern Hemisphere) or drops (Southern Hemisphere) near the pole. AOD is lower than the global mean equatorward of the injection sites, with the lowest point being at the equator in all models except MIROC, in which the local minimum is closer to 10° N. Models disagree on the ratio of polar AOD to tropical AOD, ranging from about 1.5 (MIROC) to about 3.5 (E3SM). The G6sulfur models have a higher AOD in the tropics, as well as a much higher uncertainty of tropical AOD; the four G6-1.5K-SAI models generally agree more strongly with each other on the shape of the AOD distribution than the G6sulfur models, even if one or both of the models that used different methods to compute AOD are excluded.

In Fig. 4, we present atmospheric sulfur lifetime and dry effective aerosol radius (Reff) for participating models. Sulfur lifetime can be computed as the ratio of the global burden to the global sink, which we estimate by assuming a quasi-steady state where the sink is approximately equal to the dominant source, a method used in previous studies of stratospheric aerosol injection e.g.,. Therefore, we compute sulfur lifetime by taking the ratio of change in global sulfur burden to sulfur injection rate (the S fraction of SO₂ injection rate; note that injection rates in Fig. 4a are approximately half those in Fig. 2a). Global sulfur burden is derived from the sulfur fractions of global SO₄ burden (CESM, E3SM), global H₂SO₄ burden (UKESM), or global aerosol burden (MIROC), assuming that differences between SSP2-4.5 and G6-1.5K-SAI are dominated by the SO₂ injections. Using this method, we report sulfur lifetimes between 1 and 1.5 years for CESM, E3SM, and MIROC, and a lifetime of approximately 8 months for UKESM (Fig. 4b). reported similar stratospheric sulfate lifetimes for single-latitude 30° N and 30° S injection in CESM2 and UKESM1, and also estimated an aerosol lifetime of less than one year for 30° N + 30° S injection in UKESM1. For comparison, reported lifetimes of about 16 months and 10 months for 23–25 and 19–20 km altitude SAI, respectively, in CESM1, and reported lifetimes of ∼ 12–13 months for SAI in a previous generation of models.

Estimates of dry effective aerosol radius, Reff, are computed using the methodology outlined by , Appendix A, and ; we describe this process, and the relevant equations, in Sect. S2 of the Supplementary Material. Using this method, we compute zonal mean Reff averaged over 2065–2084 (using monthly output; assuming a relatively even longitudinal distribution of aerosols after 30 years of injection, we do not weight longitudinally by, for example, surface area density or mass) for G6-1.5K-SAI for participating models (Fig. 4c–f). Reff peaks in the lower stratosphere for all models, with some models (E3SM and MIROC) showing a higher variance in latitudinal size distribution. Models report a peak Reff ranging from 0.3–0.35 µm (UKESM) to 0.4–0.45 µm (CESM). CESM has the largest Reff ( also reported CESM2 having larger Reff than the other two other participating models in that study with sectional aerosol representation); CESM has injection rates which are closest to the median of the four participating models, but also has a higher aerosol lifetime than the models which inject more. UKESM, which has the smallest peak Reff, has the highest injection rates but also a much shorter lifetime than the other models. As aerosols grow beyond the optimal radius for scattering sunlight for a given wavelength estimate the optimal wet radius for sulfate to be 0.30 µm, they will reflect sunlight less efficiently and cool the planet less efficiently; larger aerosols will, in general, also sediment out faster. CESM, which has the largest peak Reff, is the only G6-1.5K-SAI model to show a prominent decrease of AOD per unit injection with time, as discussed above (Fig. 2d). Note that for MIROC, which assumes a single-mode size distribution for radiative calculations, the actual aerosol size distribution affects transport but not radiative transfer. For comparison, online-calculated wet effective aerosol radius is only available as model output for CESM and MIROC, and is included in the Supplementary Material (Sect. S2).

In Figs. 5 and 6, we present maps of near-surface air temperature changes for SSP2-4.5 and G6-1.5K-SAI. Under SSP2-4.5, global warming (left column) increases surface temperatures nearly everywhere in all models, with the exceptions being the Labrador Sea in CESM and MIROC (indicative of a slowing down of the Atlantic Meridional Overturning Circulation, or AMOC, e.g., ); and parts of the Southern Ocean in MIROC. Arctic amplification (stronger warming in the Arctic relative to the rest of the world), a robust feature across observations and model simulations of warming , is observed in all four models, and Antarctic amplification is observed in CESM, E3SM, and UKESM in SSP2-4.5. Outside of the Arctic, warming over land tends to be higher than warming over the ocean in all four models.

In all four models, G6-1.5K-SAI cools the planet everywhere or nearly everywhere (middle column). Cooling tends to be strongest at the poles, but there is still statistically significant cooling in the tropics despite the AOD minimums at or near the equator. Cooling is stronger over land than over the ocean; this has been attributed to the smaller heat capacity of air compared to water in SAI simulations , but could also represent the preventing of increased warming over land under global warming, which is a robust result in model simulations and has been attributed to a number of hydrological and lapse rate feedbacks .

The only location where G6-1.5K-SAI induces warming is parts of the Southern Ocean in MIROC; however, these are the same locations as where global warming cools the surface in the same model, meaning the temperature changes are likely indicative of circulation changes under global warming which are mitigated by the intervention.

Because SAI cools the planet by a different mechanism (reflected solar radiation) than that by which climate change warms it (trapped infrared radiation), SAI does not revert a warmed climate to its pre-warmed condition; rather, the combination of greenhouse gas effect and SAI creates a new, unique climate state. The right column of Fig. 5 compares this new state with the “reference period” of SSP2-4.5 from which GMSAT targets were derived. In all four models, under SSP2-4.5, the NH warms more than the SH due to a combination of Arctic amplification and increased warming over land relative to the ocean; G6-1.5K-SAI offsets both of these mechanisms by cooling the Arctic and land surface most strongly, but the extent to which hemispherically symmetric injection at 30° latitude over- or under-compensates for each is model-dependent. By design, the reference period and the last 20 years of G6-1.5K-SAI have the same global average temperature, and so any overcooling (blue) in one part of the world must be balanced by residual warming or undercooling (red) in another. In MIROC and UKESM, most of the tropics and midlatitudes are overcooled by <1 °C, with residual warming at one or both poles. CESM overcools most of the Northern Hemisphere while undercooling most of the Southern Hemisphere. E3SM has the smallest temperature differences between these two periods, with no statistically significant difference across the highest portion of the planet out of the four models.

The G6-1.5K-SAI results for CESM2 and UKESM1.1 are consistent with those of several previous SAI simulations in CESM2 and UKESM1, respectively. Under the ARISE-SAI-1.5 protocol, which manages for the same GMSAT (T0) in the same scenario while also managing the interhemispheric (T1) and equator-to-pole (T2) temperature gradients, CESM2 restores reference-period surface temperatures nearly everywhere by injecting mostly (∼ 60 %) at 15° S, with additional injection (∼ 20 % each) at 15° N and 30° S . This implies that equal injections in both hemispheres would overcool the NH, consistent with G6-1.5K-SAI; additionally, G6sulfur (equatorial injection) in CESM2 overcools the NH and undercools the SH, with little residual Arctic warming . In UKESM1, residual temperature patterns under ARISE-SAI-1.5 (mostly 30° N and 30° S injection, but with increasing 15° N injection the last 15 years) are very similar to UKESM1.1 under G6-1.5K-SAI . G6controller, which aims to reduce T0, T1, and T2 in the SSP5-8.5 scenario to SSP2-4.5 levels, also bears some similarities; the controller converges to mostly (>80 %) injection at 30° N and 30° S, with significant residual Arctic amplification .

Figure 6 presents a comparison of the G6-1.5K-SAI and G6sulfur multi-model means. Unlike G6-1.5K-SAI, which uses a single warming scenario (SSP2-4.5) and cools the planet to maintain a fixed temperature, G6sulfur cools the planet to reduce temperatures from a high-warming scenario (SSP5-8.5) to match those of a moderate-warming scenario (SSP2-4.5). For a comparison period, we choose 2070–2089 for G6sulfur, in which the models average approximately 1.4 °C of cooling, the same as the G6-1.5K-SAI models in 2065–2084; therefore, the analogous “reference period” (period #1 in Fig. 6) with the same global mean temperature as the intervention is SSP2-4.5 2070–2089, the analogous “warmed world” with 1.4 °C of warming relative to the reference period (period #2) is SSP5-8.5 2070–2089, and the analogous “new climate state” with 1.4 °C of warming and 1.4 °C of cooling (period #3) is G6sulfur 2070–2089.

Comparing the patterns of warming in the two scenarios (left column), the G6-1.5K-SAI warming scenario (SSP2-4.5 relative to baseline) has more warming in the Arctic, while the G6sulfur warming scenario (SSP5-8.5 relative to SSP2-4.5) has more warming over land outside of the Arctic. Both scenarios have increased warming over land relative to the oceans, and more warming in the Arctic than the Antarctic. Both interventions (middle column) cool the land more than the ocean, and cool the NH more than the SH; G6-1.5K-SAI cools the Arctic and land in the NH more strongly than G6sulfur, and G6sulfur cools land in the SH more strongly than G6-1.5K-SAI. Both strategies cool Antarctica comparably to each other. Relative to their respective warming scenarios, G6sulfur has higher residual Arctic and NH midlatitude warming, and more G6sulfur models agree on where there is residual polar warming in both hemispheres (right column). Note that the residuals are functions of both the injection strategy and the scenario in which it is used; for example, G6-1.5K-SAI has more residual cooling over land in the NH, and while G6-1.5K-SAI does cool the NH land more strongly than G6sulfur, there is also less warming over land to offset in the former’s warming scenario than the latter’s.

In the G6-1.5K-SAI warming scenario (SSP2-4.5), there is a pattern of increased warming in the eastern tropical pacific, indicative of a positive ENSO (El Niño Southern Oscillation) response. This pattern is absent in the G6sulfur warming scenario (SSP5-8.5 minus SSP2-4.5). Under G6-1.5K-SAI, there is a cooling pattern in this region, but smaller in magnitude and size than the warming, leaving a pattern of warming in the residual. There has been some investigation as to whether and how SAI could affect ENSO; analyzed the GeoMIP G1, G2, G3, and G4 simulations and found little change to ENSO frequency or amplitude under SRM. A study of global teleconnection patterns in CESM found little change in ENSO behavior under RCP8.5 or GLENS in CESM1, but modest changes under SSP5-8.5 in CESM2 which were offset by SAI , and another study found that under ARISE-SAI-1.5, some seasonal predictability is maintained relative to SSP2-4.5 in regions affected by ENSO teleconnections . It is plausible that, in this case, the temperature patterns visible in Fig. 6 reflect modulation of ENSO (or lack thereof) under SAI; however, we leave in-depth investigation of ENSO behavior to a future study.

In Fig. 7, we present maps of precipitation changes between the same time periods as above. Under global warming, precipitation is expected to increase as the warmer atmosphere’s capacity to hold water increases in accordance with the Clausius-Clayperon relation. Under SSP2-4.5 (left column), changes in all four models share common characteristics: precipitation increases in the mid- and high latitudes in both hemispheres, statistically insignificant changes in much of the subtropics and extratropics, and the largest changes near the equator, especially in the Pacific. The tropical changes generally reflect meridional shifts of precipitation, although the models disagree on the direction of the change; CESM shows a southward shift, E3SM shows a northward shift, MIROC shows a convergence at the equator from both directions, and UKESM shows only an increase without a decrease of comparable size nearby.

The impacts of G6-1.5K-SAI (middle column) are also most prominent in the tropics, midlatitudes, and poles, with all four models having the most area without any statistically significant change in the subtropics and extratropics. Cooling the planet reduces precipitation in the mid- and high latitudes in both hemispheres in all four models, offsetting the warming-induced increases. In all four models, these decreases under G6-1.5K-SAI are of comparable magnitude to the increases under SSP2-4.5. As a result, differences in the residual (right column) are statistically insignificant over large portions of the Earth’s surface outside of the tropics in all four models. As with SSP2-4.5, the largest precipitation changes under G6-1.5K-SAI are found near the equator, but they are harder to characterize; some changes due to global warming are mitigated, others are exacerbated, and sometimes new changes are introduced.

Precipitation changes in CESM2 under G6-1.5K-SAI follow a similar pattern to those under ARISE-SAI-1.5 seeFig. 6, with comparable midlatitude decreases offsetting the increases under SSP2-4.5 and a similarly-structured shift in the tropical Pacific, and precipitation responses in UKESM1.1 under G6-1.5K-SAI are generally more similar to UKESM1 G6controller than to UKESM1 G6sulfur seeFig. 5. However, differences in residuals in all cases are likely functions of both differences in scenario and time period, as well as differences in strategy (see G6-1.5K-SAI and G6sulfur comparison below).

Figure 8 presents a multi-model comparison of G6-1.5K-SAI and G6sulfur, averaging over the same time periods as in Fig. 6 when both experiments cool the planet by approximately 1.4 °C. As with temperature differences, residual precipitation differences relative to the reference in each experiment (right column) must take into account differences in both the warming scenario (SSP2-4.5 for G6-1.5K-SAI, and SSP5-8.5 for G6sulfur) and the impacts of the intervention itself. In the G6sulfur warming scenario, on average, precipitation changes in the tropical Pacific ocean are similar to the warming scenario of G6-1.5K-SAI, but changes in the tropics outside of the Pacific are more pronounced. G6sulfur reduces precipitation in the tropics by a larger degree than G6-1.5K-SAI, both in the Pacific and elsewhere; G6sulfur also increases precipitation in more parts of the tropics and subtropics than G6-1.5K-SAI. Compared to the changes under their respective warming scenarios, G6-1.5K-SAI and G6sulfur both have the largest residual differences in the tropical Pacific; however, G6sulfur has a stronger net drying effect in this region, and the residual drying is also prominent over land in South America, Malaysia, and central Africa, whereas the residual precipitation differences of G6-1.5K-SAI are more concentrated in the ocean. Residual drying in the midlatitudes in both hemispheres is also more pronounced for G6sulfur than for G6-1.5K-SAI.

In the Supplementary Material (Sect. S3), we provide additional diagnostics characterizing the intertropical convergence zone (ITCZ). The impacts of SAI on the ITCZ are frequently cited as an important metric; it is known that preferrentially forcing one hemisphere would be expected to shift the ITCZ towards the opposite hemisphere e.g.,, and the choice to manage T1 by was motivated by the goal of limiting disruption to tropical precipitation. Here, we compute the location of the ITCZ as the centroid of zonal mean tropical precipitation between 20° N and 20° S latitude e.g.,after ; we also compute the strength of the ITCZ as the maximum zonal mean rainfall and the width of the ITCZ as the size of the latitude band containing the zonal maximum where precipitation exceeds 5 mm d⁻¹ after. While we refrain from a detailed analysis, changes in ITCZ location for G6sulfur and G6-1.5K-SAI models are small relative to variability, and increases in ITCZ strength and width under global warming tend to be mitigated under G6-1.5K-SAI and mitigated or overcompensated by G6sulfur.

In Fig. 9, we present a comparison of hydrological sensitivity for G6-1.5K-SAI and G6sulfur. It is commonly seen in SRM simulations that cooling the entire planet reduces precipitation by a greater extent than it increased under the amount of warming to be offset. In the G1 experiment (uniform solar dimming to offset the forcing increase from an abrupt quadrupling of CO₂ concentrations), precipitation decreased by 2 %–6 % relative to preindustrial control in both CMIP5 and CMIP6 models ; in SAI simulations, precipitation decreases relative to the reference in GLENS , ARISE-SAI-1.5 in both models , UKESM1’s G6controller , and G6sulfur in all participating models (Fig. 9); and there is residual warming in a CESM2 SAI experiment controlling for global mean precipitation . In the G6-1.5K-SAI simulations, global mean precipitation also decreases relative to that of the reference period (SSP2-4.5 2020–2039) in all participating models.

The residual decrease in precipitation for G6-1.5K-SAI is smaller than for G6sulfur; this result is a combination of both the difference in scenario (larger precipitation increases in the warming scenario of G6-1.5K-SAI) and differences in the effects of the intervention (smaller absolute decreases in precipitation for G6-1.5K-SAI). Between 2020–2039 and 2065–2084 in the SSP2-4.5 scenario for G6-1.5K-SAI-participating models, precipitation increases by 0.09 ± 0.03 mm d⁻¹; in G6sulfur-participating models, in SSP5-8.5 2070–2089, relative to SSP2-4.5 2070–2089, precipitation increases by only 0.05 ± 0.02 mm d⁻¹. G6-1.5K-SAI models, on average, reduce precipitation by 0.11 ± 0.04 mm d⁻¹ during the 2065–2084 period, when they cool the planet by ∼ 1.4 °C; during the period when G6sulfur models cool the planet by ∼ 1.4 °C (2070–2089), they reduce precipitation by 0.14 ± 0.03 mm d⁻¹. Differences in hydrological sensitivity between G6sulfur and G6-1.5K-SAI could be due to the difference in scenario as well as the difference in strategy; analyzed a similar pair of experiments (G6sulfur and ARISE-SAI-1.5) and suggest that the choice of stabilized vs. transient temperature target can also affect hydrological response to SAI.