Evaluation of impacts of SCI requires a set of non-SCI reference simulations to enable comparison of impacts with and without SAI. As motivated by MacMartin et al. (2022), we use the moderate Shared Socioeconomic Pathway scenario of SSP2–4.5 for our simulations, which more closely captures current policy scenarios compared to higher emission scenarios such as SSP5–8.5 (Burgess et al., 2020). SSP2–4.5, which marks a continuation of the Representative Concentration Pathway 4.5 (RCP4.5) scenario, is a “middle-of-the-road”, intermediate mitigation scenario in which “the world follows a path in which social, economic, and technological trends do not shift markedly from historical patterns” (O'Neill et al., 2017), representing the medium range of future forcing pathways (O'Neill et al., 2016).

The ARISE-SAI simulations are designed to simulate a plausible implementation scenario of SCI using SAI for evaluation of potential climate intervention risks and impacts. MacMartin et al. (2022) described in detail the need for various scenarios to evaluate impacts of SCI and five dimensions of SCI deployment options which include the background climate change scenario, desired target of cooling, start date of deployment, how cooling is achieved, and other factors that could affect decisions. The proposed default ARISE-SAI protocols closely follow the recommended scenario choices described in MacMartin et al. (2022) and describe details of implementation in Earth system models, although different choices can be made in the future to expand the simulation set. In particular, the proposed ARISE-SAI simulations utilize a moderate emission scenario, SSP2–4.5 (O'Neill et al., 2016), and cool the Earth to a global mean temperature target (TT) above pre-industrial levels denoted in the specific name of the simulations (e.g., ARISE-SAI-TT). For example, ARISE-SAI-1.5 and ARISE-SAI-1.0 simulations aim to maintain global surface temperatures at ∼1.5 and ∼1.0 ∘C above pre-industrial levels, respectively.

The protocol in the first ARISE-SAI simulations (without a delayed start) simulates deployment beginning in 2035 after the global surface temperature reaches ∼1.5 ∘C above pre-industrial levels, which is the target proposed in the 2015 Paris Agreement and described by the IPCC as an important threshold for climate safety (IPCC, 2018). Simulations are carried out for 35 years (2035–2069), which is sufficient to consider both a transition period of ∼10 years and a quasi-equilibrium of at least 20 years after the controller converges. Minimum recommended ensemble size is three, although more members will allow for more thorough evaluation of impacts on variability.

The first ARISE-SAI simulations, ARISE-SAI-1.5 presented here, aim to keep the global mean temperature at ∼1.5 ∘C above pre-industrial levels. There is uncertainty among Earth system models with regard to when Earth's global mean surface temperature (T0) will reach 1.5 ∘C above pre-industrial levels. The recent Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report (AR6) (IPCC, 2021) finds that 1.5 ∘C over pre-industrial will very likely be exceeded in the near term (2021–2040) under the very high greenhouse gas (GHG) emission scenario (SSP5–8.5) and is likely to be exceeded under the intermediate and high GHG emissions scenarios (SSP2–4.5 and SSP3–7.0). The IPCC AR6 defines 1.5 ∘C as the time at which T0 will reach 0.65 ∘C above the historical reference period of 1995–2014. The T0 between 1995 and 2014 is 0.85 ∘C above the pre-industrial (PI) value defined as the 1850–1900 average in the observational record. Using 31 global models, Tebaldi et al. (2021) found that the average across models of when 1.5 ∘C will be reached is 2028 under the SSP2–4.5 scenario (using 1995–2014 as 0.84 ∘C rather than 0.85 ∘C above PI), but with considerable variation across models. To simplify future model intercomparisons, we choose the time period of 2020–2039 (or ∼2030 levels) as our reference period of when T0 is ∼1.5 ∘C above PI values and make that the target T0 in the ARISE-SAI-1.5 climate intervention simulations. We acknowledge that different climate models with different baseline temperatures and rates of warming might have different time periods in which they reach 1.5. Nonetheless, we recommend that the best way to achieve a meaningful and easy comparison between different models would be to always use the model's 2020–2039 SSP2–4.5 period as a baseline over which to calculate the targets of ARISE-SAI-1.5 simulations. This way, the reference period is the same between models and the 2035 start date remains meaningful in every case.

In addition to keeping T0, the ARISE-SAI simulations aim to keep the north–south temperature gradient (T1) and Equator-to-pole temperature gradient (T2) to those corresponding to the temperature target. This is achieved by utilizing a “controller” algorithm (MacMartin et al., 2014; Kravitz et al., 2017) that specifies the amount of SO2 injection. This approach was used in GLENS and the simulations presented in Tilmes et al. (2020). The controller algorithm is freely available as described in the “Code availability” section. Sulfur dioxide injections in the ARISE-SAI simulations are placed at four injection locations (15∘ S, 15∘ N, 30∘ S, 30∘ N) into one grid box at ∼21.5 km altitude. The injection latitudes are the same as used in GLENS and in previous studies examining the model's responses to single-point SO2 injections (Tilmes et al., 2017; Richter et al., 2017). These four injection locations are sufficient to independently control the targets that we are trying to achieve (Kravitz et al., 2017). These four injection locations have also been demonstrated to be sufficient to produce the optical depth patterns that independently control the targets that we are trying to achieve in various versions of CESM(WACCM) (MacMartin et al., 2017; Zhang et al., 2022; MacMartin et al., 2022). The prescribed injection altitude is estimated to be achievable by existing aircraft technologies that could be adapted for climate intervention use (Bingaman et al., 2020). After each year of simulation, the algorithm calculates the global mean temperature (T0), north–south temperature gradient (T1), and Equator-to-pole temperature gradient (T2), and based on the deviation from the goal, it specifies the annual values of injections at the four locations for the subsequent year. T1 and T2 were defined in Kravitz et al. (2017) in Eq. (1).

Comprehensive monthly output as well as high-frequency output for analysis of high-impact events (described in detail in the “Data records” section) are needed for analysis of SCI impacts on the Earth system. Acknowledging limitations of various modeling centers, we recommended a minimum set of monthly mean output fields in Table A1 in the “Data records” section and include the full comprehensive output list that was created with the CESM2(WACCM) simulations based on input from the broader community. All model output for the simulations should be provided in NetCDF format. All variables should be in time series format, with one variable per file. Three-dimensional atmospheric output should be on the original model levels or on standard CMIP6 levels. For monthly atmospheric output, information on aerosol microphysics (which is not a standard CMIP6 output) is also very relevant for diagnostics of the aerosols' behavior under SAI; for instance, CESM2(WACCM6) includes as standard output the mass and number concentration for all aerosol modes and the aerosol effective radius. Other modeling centers should consider providing this (model specific) information as well. In addition, higher-frequency (daily averaged, 3-hourly averaged, 3-hourly instantaneous, and 1-hourly mean) output is desired for the atmospheric model that will enable analysis of extreme events (e.g., Tye et al., 2022). The atmospheric output at various time frequencies is described in Appendix A in Tables A2–A5. Daily averaged output of land model variables is shown in Tables A6 and A7, whereas 6-hourly output from the land model is listed in Table A8. Tables A9 and A10 show the daily output from the ocean and sea ice models, respectively. The table captions describe which output is specific to ARISE-SAI-1.5 and the five new SSP2–4.5 CESM2(WACCM6) ensemble members and which is common to all simulations. An online table showing all the output fields for the simulations, along with their description and units, is at https://www.cgd.ucar.edu/ccr/strandwg/WACCM6-TSMLT-SSP245/ (last access: 11 November 2022).

The ARISE-SAI-1.5 simulations described above are likely to be most relevant to policy makers, and hence reproduction of the experiments in multiple models is desired. ARISE-SAI simulations are already being performed with the UKESM. ARISE-SAI-1.0 simulations and ARISE-SAI-1.5-2045, with the start of intervention delayed by 10 years, are in progress with CESM2(WACCM). A subset of simulations describing these different initial conditions and targets is discussed in MacMartin et al. (2022) using a slightly more simplified version of CESM2(WACCM6).

CESM2(WACCM6) is the most comprehensive version of the NCAR whole-atmosphere ESM and is described in detail in Gettelman et al. (2019) and Danabasoglu et al. (2020). CESM2(WACCM6) was used to contribute climate change projection simulations to the Coupled Model Intercomparison Project Phase 6 (CMIP6) (Eyring et al., 2016). CESM2(WACCM6) is a fully coupled ESM with prognostic atmosphere, land, ocean, sea ice, land ice, and river and wave components. The atmospheric model, WACCM6, uses a finite-volume dynamical core with a horizontal resolution of 1.25∘ longitude by 0.9∘ latitude. WACCM6 includes 70 vertical levels with a model top at 4.5×106 hPa (∼140 km). Tropospheric physics in WACCM6 are the same as in the lower top configuration, the Community Atmosphere Model version 6 (CAM6). CESM2(WACCM6) includes a parameterization of non-orographic waves which follows Richter et al. (2010) with changes to tunable parameters described in Gettleman et al. (2019). Parameterized gravity waves are a substantial driver of the quasi-biennial oscillation (QBO), which is internally generated in CESM2(WACCM6). CESM2(WACCM6) includes prognostic aerosols which are represented using the Modal Aerosol Model version 4 (MAM4) as described in Liu et al. (2016). This includes four modes, only three of which are used for sulfate: Aitken, accumulation, and coarse mode. In the stratosphere, CESM(WACCM6) includes a comprehensive interactive sulfur cycle, as described, for instance, in Mills et al. (2016); this allows for SO2 oxidation (with interactive OH concentration) and subsequent nucleation and coagulation of H2SO4 into sulfate aerosol (allowing for inter-mode transfer), which are then removed from the stratosphere through gravitational settling and large-scale circulation. A more in-depth analysis of the size distribution and vertical distribution of sulfate aerosols under SO2 injections has been performed in Visioni et al. (2022) (for single-point injections at the same latitudes and altitudes as those described in these simulations), also compared with results from other models with similar aerosol microphysics (UKESM1 and GISS), highlighting that in CESM2(WACCM6) the produced stratospheric aerosol is mainly found in the coarse mode. CESM2(WACCM6) also includes a comprehensive chemistry module with interactive tropospheric, stratospheric, mesospheric, and lower thermospheric chemistry (TSMLT) with 228 prognostic chemical species, as described in detail in Gettleman et al. (2019).

The ocean model in CESM2(WACCM6) is based on the Parallel Ocean Program version 2 (POP2; Smith et al., 2010; Danabasoglu et al., 2012, 2020). The horizontal resolution of POP2 is uniform in the zonal direction (1.125∘) and varies from 0.64∘ (occurring in the Northern Hemisphere) to 0.27∘ at the Equator. The ocean biogeochemistry is represented using the Marine Biogeochemistry Library (MARBL), which is an updated implementation of the Biochemistry Elemental Cycle (Moore et al., 2002, 2004, 2013). CESM2 uses version 3.14 of the NOAA WaveWatch-III ocean surface wave prediction model (Tolman, 2009). Sea ice in CESM2(WACCM6) is represented using CICE version 5.1.2 (CICE5; Hunke et al., 2015) and uses the same horizontal grid as POP2.

CESM2(WACCM6) uses the Community Land Model version 5 (CLM5) (Lawrence et al., 2019). CLM5 includes a global crop model that treats planting, harvest, grain fill, and grain yields for six crop types (Levis et al., 2018), a new fire model (F. Li et al., 2013; Li and Lawrence, 2017), multiple urban classes and an updated urban energy model (Oleson and Feddema, 2019), and improved representation of plant dynamics. The river transport model used is the Model for Scale Adaptive River Transport (MOSART; H. Y. Li et al., 2013).

A five-member reference ensemble with CESM2(WACCM6) and the SSP2–4.5 scenario was carried out as part of the CMIP6 project for the years 2015–2100. Surface temperature evolution and equilibrium climate sensitivity in these simulations are described in detail in Meehl et al. (2020). We carried out an additional five-member ensemble of these simulations from the years 2015–2069 with augmented high-frequency output for high-impact event analysis, as well as additional output for the land model to match the SCI simulations (Richter and Visioni, 2022a). The additional five-member ensemble was branched from the three existing historical CESM2(WACCM6) simulations in the same manner as the first five-member ensemble, but with an addition of small temperature perturbations for each ensemble member ([6, 7, 8, 9, 10] × 10-14 K, respectively) at the first model time step. CESM2 ranks highly against other CMIP6 models in the ability to represent large-scale circulations and key features of tropospheric climate over the historical time period (e.g., Simpson et al., 2020; DuVivier et al., 2020; Coburn and Pryor, 2021).

In CESM2(WACCM6) SO2 injections were placed at 180∘ longitude and bounded by two pressure interfaces: 47.1 and 39.3 hPa (approximate geometric altitude at grid box midpoint of 21.6 km). Based on the 2020–2039 mean of the SSP2–4.5 simulations with CESM2(WACCM6), the surface temperature targets for the ARISE-SAI-1.5 ensemble for T0, T1, and T2 are 288.64, 0.8767, and -5.89 K, respectively. As noted in Sect. 2.3, we recommend that T0, T1, and T2 targets for other models reproducing ARISE-SAI-1.5 simulations be based on the 2020–2039 average from their SSP2–4.5 simulations.

The first five members of ARISE-SAI-1.5 simulations were initialized in 2035 from the first five members (001 to 005) of the SSP2–4.5 simulations carried out with CESM2(WACCM6); hence, all had different initial ocean, sea ice, land, and atmospheric initial conditions on 1 January 2035. Similarly to the SSP2–4.5 simulations, subsequent ensemble members (006 through 010) were initialized from the same initial conditions as members 001 through 005, respectively, with an addition of a small temperature perturbation to the atmospheric initial condition to create ensemble spread (Richter and Visioni, 2022b).

The amount of SO2 injection in the ARISE-SAI-1.5 simulations chosen by the controller algorithm is shown in Fig. 1. The majority of SO2 is injected at 15∘ S, with an approximate linear increase from 0.5 Tg SO2 per year in 2035 to 6 Tg SO2 per year in 2069. SO2 injections at 30∘ S and 15∘ N are about 1/3 of that injected at 15∘ S. Throughout all the ARISE-SAI-1.5 simulations, the amount of SO2 injection at 30∘ N is very small at less than 0.5 Tg SO2 per year, diminishing to nearly zero by the end of the simulations. The distribution of SO2 across the four injection latitudes in ARISE-SAI-1.5 is very different from that in GLENS (Tilmes et al., 2018) despite having the same goals for the controller. In GLENS, the majority of SO2 was injected at 30∘ S and 30∘ N, with a significant amount at 15∘ N and almost none at 15∘ S; that is, GLENS required more injection in the Northern Hemisphere than the Southern Hemisphere in order to maintain the interhemispheric temperature gradient T1, whereas ARISE-SAI-1.5 requires more injection in the Southern Hemisphere to maintain T1. GLENS also required more SO2 injection at 30∘ N, 30∘ S to maintain T2 than is required in ARISE-SAI-1.5. It is unclear at this time how much of this difference is a result of the different model version and how much is a result of changes in the forcing between RCP8.5 and SSP2–4.5.

Injection of sulfur dioxide into the stratosphere results in the formation of sulfate aerosols, which are transported by the stratospheric Brewer–Dobson circulation (Andrews et al., 1987; Tilmes et al., 2017). The dominance of SO2 injections at 15∘ S in ARISE-SAI-1.5 results in a stratospheric sulfate (SO4) increase that primarily occurs in the Southern Hemisphere, with the majority of SO4 concentrated near the primary injection location (Fig. 2a and b). Averaged over the 2035–2054 period, there is a peak SO4 increase of 25 mg S kg-1 air (Fig. 2a) relative to the 2020–2039 mean, and averaged over 2050–2069 an SO4 increase of 48 mg S kg-1 air is found near 15∘ S at 40 hPa (Fig. 2b). The zonally averaged latitudinal distribution of the increase in the column of SO4 is shown in Fig. 2c and d; both show the strong hemispheric asymmetry as well as a double peak at around 15∘ S and one near 50∘ S. The peak near 15∘ S is due to the predominant location of the injection and matches the peak in concentration; the latter is due to the largest vertical stratospheric layer over which SO4 is spread out (between 10 and 22 km) compared to the layer in the tropical stratosphere (between 18 and 26 km). Integrated over 20-year periods of ARISE-SAI-1.5 simulations, there is little difference in the latitudinal distribution of column SO4 between the various ensemble members, but amplitude differences of up to 15 % exist (not shown), reflecting variability in the amount of SO2 injection at each location and small differences in the stratospheric circulation.

Global mean surface temperature, the interhemispheric temperature gradient, and Equator-to-pole temperature gradients for the SSP2–4.5 and ARISE-SAI-1.5 simulations are shown in Fig. 3. There is a notable difference in behavior of T1 and T2 in the SSP2–4.5 simulations compared to the RCP8.5 simulations with CESM1(WACCM) (not shown). In the CESM1(WACCM) simulations with RCP8.5, T1 and T2 increased steadily with time of simulation, reaching a change in T1 of nearly 0.45 K and a T2 change of 0.3 K by 2070 relative to the ∼2020–2039 mean (Tilmes et al., 2018). In contrast, T1 and T2 in the SSP2–4.5 simulation increase much more slowly by less than 0.05 K for T1 and less than 0.1 K for T2 between the reference period (2020–2039) and 2070. The more moderate (SSP2–4.5) emission scenario used in the CESM2(WACCM6) control simulations partially explains the slower increase in T1 and T2 with time, but not all. Simulations with CESM2(WACCM6) and SSP5–8.5 scenarios also show a much slower increase in T1 and T2 compared to CESM1(WACCM) with RCP8.5. Differing modeling physics, in particular cloud feedbacks, between CESM1 and CESM2 are key differences that could lead to the differences in projected spatial patterns of surface warming between the two model configurations, as well as changes in the Atlantic Meridional Overturning Circulation as discussed in Tilmes et al. (2020). Additional simulations with CESM2 and RCP emissions have been performed to understand the relative role of differences in forcing and differences in model physics in projected spatial patterns of global mean temperature and other variables between CESM1 and CESM2. A detailed discussion of the reasons behind the model dependence on injection strategy in GLENS, CESM1(WACCM), and ARISE-SAI-1.5, CESM2(WACCM6) simulations can be found in Fasullo and Richter (2022). They show that the main contributors to the differences are rapid adjustment of clouds and rainfall to elevated levels of carbon dioxide, dynamical responses in the Atlantic Meridional Overturning Circulation (AMOC), and differences in future climate forcing scenarios.

The differences between the projected surface temperature patterns in CESM2 compared to CESM1 have implications for climate intervention. Since the changes in T1 and T2 targets differ between the CESM1(WACCM) and CESM2(WACCM6) future simulations, the controller selects different SO2 injection locations to best counteract these changes. Injections needed to offset increasing T1 and T2 in CESM1(WACCM) required primarily injections at 30∘ S and 30∘ N, whereas for a small change in T1 and T2 relative to the 2020–2039 period in CESM2(WACCM6), SSP2–4.5 requires injections primarily at 30∘ S. The SO2 injections applied in ARISE-SAI-1.5 do a very good job at keeping the global mean temperature, T1, and T2 at the target levels. This is demonstrated by the blue lines in Fig. 2. There is a fair amount of variability among the individual ensemble members (thin light blue lines) in their ability to meet the global mean T1 and T2 targets; however, the ensemble mean (thick blue line) shows very good agreement between these variables and their target values.

Figure 4 shows the ensemble and annual mean surface temperature changes for two time periods, 2035–2054 and 2050–2069, during the SSP2–4.5 and ARISE-SAI-1.5 simulations relative to the 2020–2039 period. Figure 4a and c show the steady increase in surface temperature with time over the majority of the globe, with the largest warming occurring in the Northern Hemisphere high latitudes. The North Atlantic is the only region of the globe that is cooling in the 21st century. This “warming hole” in the North Atlantic is a feature of several recent-generation Earth system models and is attributed to the AMOC (Drijfhout et al., 2012; Chemke et al., 2020; Keil et al., 2020). Specifically, in a warming climate with a reduction in deepwater formation, the AMOC weakens. This results in less heat transport into the northern North Atlantic, producing cooler temperatures that oppose the anticipated effects of global warming. Figure 4b and d demonstrate the success of the SAI strategy in keeping the global temperatures near the 2020–2039 average, or at ∼1.5 K above pre-industrial values. In ARISE-SAI-1.5, near-surface annual mean temperature throughout the entire simulation is within 0.5 K of that goal over the majority of the globe. The largest exception to that is the North Atlantic warming hole, where surface temperatures remain cooler relative to the northern North Atlantic than in the present day; while AMOC strength is partially recovered under SAI relative to SSP2–4.5, it is not fully restored back to present-day conditions. In addition, in the ensemble mean, ARISE-SAI-1.5 simulations show residual warming over North America, as well as over eastern South Pacific Ocean (off the coast of South America) and in parts of Antarctica compared to the 2020–2039 period. Residual changes relative to the target period from the application of SAI are expected, as SAI cannot perfectly reverse the effects of increasing greenhouse gases.

The precipitation changes in SSP2–4.5 and ARISE-SAI-1.5 simulations for the same time periods examined for surface temperature changes are shown in Figs. 5 and 6. Consistent with prior similar studies, SSP2–4.5 simulations primarily show an increase in precipitation in a warming climate, with the largest increases along the equatorial Pacific Ocean and a strong drying region northward of that (Figs. 5, 6a and c). In ARISE-SAI-1.5, consistent with previous studies (Kravitz et al., 2017; Lee et al., 2020), restoring global mean temperature is associated with an overall decrease in annual mean precipitation (Fig. 5); however, regionally both increases and decreases occur. In ARISE-SAI-1.5, the increased precipitation across the equatorial Pacific seen in SSP2–4.5 decreases in magnitude but is still a persistent feature. ARISE-SAI-1.5 also shows drying north and south of that region as well as intensified drying over northern South America, South Africa, the Indian Ocean south of the Equator, and northernmost Australia. The Indian Ocean north of the Equator and India are projected to be wetter in ARISE-SAI-1.5 compared to the 2020–2039 period of SSP2–4.5.