Impact of stratospheric aerosol intervention geoengineering on surface air temperature in China: a surface energy budget perspective

Abstract. Stratospheric aerosol intervention (SAI) geoengineering is a rapid, effective, and promising means to counteract anthropogenic global warming, but the climate response to SAI, with great regional disparities, remains uncertain. In this study, we use Geoengineering Model Intercomparison Project G4 experiment simulations from three models (HadGEM2-ES, MIROC-ESM, and MIROC-ESM-CHEM) that offset anthropogenic forcing under medium-low emissions (RCP4.5) by injecting a certain amount of SO2 into the stratosphere every year, to investigate the surface air temperature response to SAI geoengineering over China. It has been shown that the SAI leads to surface cooling over China over the last 40 years of injection simulation (2030–2069), which varies among models, regions and seasons. The spatial pattern of SAI-induced temperature changes over China is mainly due to net surface shortwave radiation changes. We find that changes in solar radiation modification strength, surface albedo, atmospheric water vapor and cloudiness affect surface shortwave radiation. In summer, the increased cloud cover in some regions reduces net surface shortwave radiation, causing strong surface cooling. In winter, both the strong cooling in all three models and the abnormal warming in MIROC-ESM are related to surface albedo changes. Our results suggest that cloud and land surface processes in models may dominate the spatial pattern of SAI-induced surface air temperature changes over China.



Introduction
The increasing anthropogenic greenhouse gas (GHG) concentrations since the industrial revolution have led to global warming.Although the international community has realized the risk of global warming and attempted to reduce GHG emissions, global GHG emissions still show a continuous increase (United Nations Environment Programme, 2020).The "2°C global temperature target" in the Paris Agreements will be unachievable if the current increasing emission trend persists (e.g., Robiou du Pont and Meinshausen 2018).Geoengineering, which aims to counteract global warming by deliberately changing the climate system, is therefore of great research interest.
Geoengineering schemes are generally classified into two major types: carbon dioxide removal (CDR) geoengineering by reducing atmospheric carbon dioxide concentration, and solar radiation modification (SRM) geoengineering by increasing planetary albedo.Solar radiation modification (SRM), which refers to a range of measures adjusting the Earth's radiative balance, is considered as an option to counteract anthropogenic global warming.Various specific techniques have been proposed to perform SRM geoengineering, such as injecting sulfate aerosols into the stratosphere (Budyko, 1977), placing shields or deflectors in space (Seifritz, 1989), brightening marine clouds (Latham, 1990), and thinning cirrus clouds (Mitchell and Finnegan, 2009).The method of injecting sulfate aerosols or their precursors into the stratosphere, also known as stratospheric aerosol intervention (SAI) geoengineering, is designed to cool the surface by using these aerosols to reflect and scatter solar radiation (Crutzen, 2006;Wigley, 2006).As a proposed scheme, SAI has attracted great attention recently due to its assumed technological feasibility (e.g., Irvine et al., 2016).SAI geoengineering is considered the most promising SRM method due to its high effectiveness, affordability, and timeliness (Shepherd et al., 2009).SRM geoengineering has not been implemented in reality because of its potential risks and immature technology.The primary means of recognizing the climate response to geoengineering is by simulating via general circulation models (GCMs).However, the results from early simulations could not be proved robust due to the differences in experimental schemes.The Geoengineering Model Intercomparison Project (GeoMIP) has been proposed to address that issue (Kravitz et al., 2011;2015).
To date, the GeoMIP has designed 12 experiments, including solar dimming, stratospheric aerosol intervention, marine cloud brightening, and cirrus thinning geoengineering in Coupled Model Intercomparison Project Phases 5 and 6 (CMIP5 and CMIP6).The GeoMIP provides detailed simulating guidelines for each model and experiment and calls for all the modeling groups worldwide to become involved and share their simulations.A total of 19 GCMs have participated in the GeoMIP to date.More detailed information is accessible from the GeoMIP website (http://climate.envsci.rutgers.edu/GeoMIP/).
Previous studies have indicated that SRM geoengineering could counteract or even reverse anthropogenic global warming and reduce sea ice melting and thermosteric sea-level rise, as well as the decreaseing the frequency and intensity of extreme temperature and precipitation events (Rasch et al., 2008;Robock et al., 2015;Irvine et al., 2016;Ji et al., 2018;Jones et al., 2018).It wouldmight also come with risks.FirstFor instance, SRM geoengineering would reduces the global mean precipitation and monsoon precipitation and slows the hydrological cycle if it is used to offset the GHG-induced global warming (Bala et al., 2008;Tilmes et al., 2013;Sun et al., 2020).SRM would not mitigate the continued ocean acidification caused by CO2 emissions (Caldeira et al., 2013).Second, SRM geoengineering might induce overcooling of the tropics and undercooling of the poles, which is related to the difference between the solar and CO2 forcings (Russotto and Ackerman, 2018).Finally, tThe sudden termination of geoengineering would lead to a more rapid increase in temperature than the nongeoengineered case (Matthews and Caldeira, 2007;Jones et al., 2013).Moreover, for SAI geoengineering, the resultant cooling and enhanced polar vortex The severity of the termination effect depends on the magnitude of geoengineering deployment.Moreover, the SAI-induced heterogeneous chemistry changes might cause stratospheric ozone depletion and thus increase ultraviolet radiation (UV) at the surface (Tilmes et al., 2008;Eastham et al., 2018).
The appropriate SRM geoengineering couldmight lead to global cooling, but its regional effects might be different mainly due to the spatially heterogeneous reduction in solar radiation.This means and benefit most regions (Irvine et al., 2019).However, it was still a concern that if SRM geoengineering was performed, some regions might face greater climatic impacts or risks than othersunder SRM forcing (Ricke et al., 2013;Kravitz et al., 2014).For example, Robock et al. (2008) indicated that the SAI-induced weakening of the Asian and African summer monsoons caused by the injected stratospheric aerosols over the Arctic would decrease cloudiness and in turn warm the surface over northern Africa and India.In addition to the effect of cloudiness, changes in atmospheric moisture and surface conditions caused by SAI also impact surface air temperature (Kashimura et al., 2017).As the largest developing country in the world, China plays an important role in combating climate change.
China's attitude to SAI is crucial to the international geoengineering research community.Considering the combined effect of the Tibetan Plateau and the East Asian monsoon, the climate over China would be strongly influenced by SAI.Large volcanic eruptions, which inject massive volcanic aerosols into the stratosphere, are considered a natural analog to SAI geoengineering (Trenberth and Dai, 2007).For instance, theThe 1815 Mt.Tambora eruption led to the "year without a summer" over China (e.g., Raible et al., 2016).However,But the volcanic eruption is not a perfect analog.This is because the sulfate aerosols from massive volcanic eruptions only last for 2-3 years, while the SAI-induced aerosols are continuously replenished for decades or centuries (Duan et al., 2019).So far, few studies have studied the temperature response to the SAI geoengineering over China has not yet been studied explicitly (Cao et al., 2015).
In this study, we investigate the impact of the SAI geoengineering on the surface air temperature over China and the underlying physical processes from a surface energy perspective.Section 2 provides a brief introduction to the experiments, model data, and decomposition method of net surface shortwave radiationair temperature change.Section 3 evaluates the ability of models to reproduce the climatological temperature over China in summer and winter.Section 4 presents the summer and winter temperature changes and associated reasons over China in response to SAI geoengineering, and we also analyze the physical processes responsible for the SAI-induced net surface shortwave radiationtemperature changes over China.Conclusions and discussion are presented in Sect. 5.

Experiments
We use the simulations in the G4 experiment from the first phase of the GeoMIP.(Kravitz et al., 2011).As a SAI-based geoengineering experiment, G4 is designed to inject SO2 into the low-level equatorial stratosphere at a consistent rate of 5 Tg per year under the background scenario of Representative Concentration Pathway 4.5 (RCP4.5)(Taylor et al., 2012).This injection rate is equivalent to a case in which the 1991 Mt. Pinatubo eruption occurred every four years (Bluth et al., 1992).The injection period is from 2020 to 2069, and then the experiment continues to run until 2089 to examine the termination effect (Jones et al., 2013).The RCP4.5 simulation for the same period is used as a baseline (non-geoengineered) state.In addition, the historical simulation for 1986-2005 is applied to evaluate the ability of the selected models to reproduce the climatology of surface air temperature over China.

Data
A total of nine12 GCMs participated in the G4 experiment, and four of them are available on the Earth System Grid Federation (ESGF), including CSIRO-Mk3L-1-2, HadGEM2-ES, MIROC-ESM, and MIROC-ESM-CHEM (Kravitz et al., 2013a).However, some models should not be considered in this study due to their known issues.For instance, CSIRO-Mk3L-1-2 runs G4 by directly reducing solar irradiance rather than injecting stratospheric aerosols; GISS-E2-R shows an inconsistency between G4/RCP4.5 and historical experiments; IPSL-CM5A-LR and NorESM1-M have errors in the longwave treatment of the sulfate aerosol; GEOSCCM and ULAQ use prescribed sea surface temperatures.Note that CSIRO-Mk3L-1-2 is not selected because the clear-sky shortwave radiation flux at the surface is not available in its outputs (Phipps et al., 2011).Simulations from the other threesix models are applied for analyses.Monthly datasets are used and calculated as the averages in summer (June-July-August, JJA) and winter (December-January-February, DJF).Considering the intermodel scatter in the temperature response to SAI in the G4 experiment, we analyze the results of each model separately (Yu et al., 2015;Ji et al., 2018).The CN05.1 observation dataset (Wu and Gao, 2013) is used to evaluate the ability of models to reproduce the climatology of temperature over China.
All the observations and model outputs are interpolated to a common grid with a mid-range horizontal resolution (2.5° longitude by 2° latitude).
A brief description of the selected models used is illustrated in Table 1.In addition to differences in the physical and chemical modules related to sulfate aerosol particles, the models have different SO2 injection treatments.For HadGEM2-ES, the CLASSIC aerosol module (Bellouin et al., 2011) used in the stratosphere makes it possible to handle the injections of SO2, allowing HadGEM2-ES to finish a complete simulation including the generation and transportation of stratospheric sulfate aerosols.The injection point is located on the equator (0° longitude), and the injection altitude ranges from 16 to 25 km.For CanESM2, the stratospheric aerosol optical depth (SAOD) caused by SAI is prescribed as a consistent value.For other models (BNU-ESM, CNRM-ESM1, MIROC-ESM and MIROC-ESM-CHEM, the SPRINTARS aerosol module mainly focuses on tropospheric aerosols.The ), the prescribed distribution of stratospheric sulfate aerosol optical depth (AOD),SAOD, according to Sato (2006), is used to drive the G4 experiment.The only difference between MIROC-ESM andBesides, MIROC-ESM-CHEM is that the latter is coupled with the CHASER atmospheric chemistry module, which can be used to calculatecalculates the surface density of sulfate aerosols by using the CHASER atmospheric chemistry module (Sudo et al., 2002;Kravitz et al., 2013a).

Decomposition method for SAI-induced shortwave radiation changes at the surfacesurface air temperature change
The surface air temperature change depends on the components of the surface energy budget, including shortwave and longwave radiation (SW and LW) and sensible and latent heat (SH and LH) (Boer, 1993).For example, the surface radiation (SW and LW) changes due to SRM geoengineering may be balanced by the surface temperature and/or nonradiative (SH and LH) flux changes (Andrew et al., 2009).Previous studies indicated that SRM geoengineering could reduce the surface SW, which was mainly compensated by the decreased LH flux (e.g., Schmidt et al., 2012).Therefore, it is important to analyze the surface SW response to SAI forcing in this study.
Surface air temperature is a widely used variable in climate studies.Change in surface air temperature is associated with three components: surface vertical energy fluxes (including radiative and heat fluxes), horizontal temperature advection, and adiabatic warming or cooling (Gong et al., 2017).In this study, the SAI-induced changes in surface temperature and surface air temperature are strongly coupled in China during 2030-2069 (the correlation coefficients are higher than 0.98 and 0.99 in summer and winter, respectively; Fig. 1).Thus, the surface vertical energy fluxes are considered to be the main factor affecting temperature change under SAI forcing.
According to the decomposition method based on the surface energy budget proposed by Lu and Cai (2009), the surface air temperature change caused by SAI can be written as: A decomposition method for the SAI-induced net surface SW changes proposed by Kashimura et al. (2017) is applied for this study.That method is based on the single-layer atmospheric model of SW transfer according to Donohoe and Battisti (2011) which assumes that the transportation processes of SW, including atmospheric reflection, atmospheric absorption, and surface reflection, are isotropic.As detailed by Kashimura et al. (2017), the upward SW at the top of the atmosphere (TOA) and the upward and downward SW at the surface at each grid point can be approximated as: where ∆ represents the difference between G4 and RCP4.5, the overbar represents the climatological value of RCP4.5, R ↓ is the downward net radiation at the surface, LH and SH are surface sensible and latent heat fluxes respectively, Q is surface heat storage, Ts is surface temperature, and σ is the Stefan-Boltzmann constant.Res represents the difference between changes in surface air temperature and surface temperature.In order to quantitatively separate the radiative effects of clouds and surface albedo, the ∆R ↓ can be decomposed as follow: where S is the downward SW at the TOA (  1)-( 3) as follows: In addition, we assume that the impact of clouds on surface albedo (α) is negligible.
The values of surface albedo are uniformly calculated using the all-sky values of SW collected at the surface in this study.Here, the flux is defined as downward positive, and the net surface SW at each grid point can be represented as a function of S, R as , A as , R cs , A cs , and α as follows: As shown in Fig. 1, the effects of SAI on the net surface SW can be divided into four parts: effects of changes in the strength of solar radiation modification (SWSRM), amount of atmospheric water vapor (SWWV), cloud cover (SWC), and surface albedo (SWSA).Here we further assume that the changes in solar radiation modification strength and water vapor amount would only lead to changes in R cs and A cs respectively, and the concentrations of other atmospheric compositions related to R cs and A cs would not be affected by SAI.The net surface SW change can therefore be decomposed as follows: where F is the fraction of solar radiation reflected by the atmosphere, and A is the fraction of absorption during solar radiation passing through the atmosphere.SWSRM and SWWV represent the effects of solar radiation scattering and atmospheric water vapor amount, respectively.Although the SW cs↓ change is not precisely equal to the sum of changes in SWSRM and SWWV due to the assumption of a single-layer model (Donohoe and Battisti, 2011), this method is effective when analyzing the surface shortwave radiation change in response to SAI (Kashimura et al., 2017).
Although the net surface SW differences between G4 and RCP4.5 are not precisely equal to the sum of SWSRM, SWWV, SWC, and SWSA changes due to the assumption of a single-layer model and the nonlinearity of Eq. ( 7), this method is effective when analyzing the net surface SW change in response to SAI geoengineering.

Evaluation of the models
The ability of the models to reproduce the surface air temperature over China is evaluated first.
As shown in Fig. 2, the spatial correlation coefficient (SCC), standard deviation (SD), and centered root-mean-square error (CRMSE) between the observation and the historical simulation for the climatological temperature over China during 1986-2005 are calculated and illustrated in a Taylor diagram (Taylor, 2001).The SCCs of the models range from 0.8885 to 0.95 (0.94 in multi-model mean) in summer and from 0.91 to 0.96 (0.96 in multi-model mean) in winter.All the SCCs are statistically significant at the 99% level, meaning that the simulated temperature is in good agreement with the observed temperature.The normalized SDs range from 0.9581 to 1.0733 in summer (0.99 in multi-model mean) and from 1.0703 to 1.1623 (1.08 in multi-model mean) in winter.This result indicates that all threeselected models overestimate the spatial variability of the winter temperature except for HadGEM2-ES in summerChina.The CRMSEs are 0.31-0.5134-0.53(0.35 in multi-model mean) for summer and 0.3332-0.46(0.31 in multi-model mean) for winter.Taken together, the simulations of summer and winter temperatures by selected models are reliable over China.
HadGEM2-ES performs better than MIROC-based models, which may be related to its finer horizontal resolution The multi-model mean results outperform most individual models for the temperature climatology over China both in summer and winter, which is consistent with previous findings (e.g., Jiang et al., 2016).
The observed spatial distributionspatterns of summer and winter temperature biasesclimatology over China between simulationsshow a general decrease from south to north, and observations are shownthe lowest values mainly occur in Fig. 3.the Tibetan Plateau (Figs. 3a, d).These features can be well reproduced by all models and their mean (Figs.3b, e).Compared to the observations,observation, the simulated temperature is generally overestimated in summer but underestimated in winter over China according to the regionally averaged values.In MIROC-ESM and MIROC-ESM-CHEM, cold biases mainly occur over the Tibetan Plateau, and In summer, warm biases mainly occur in Xinjiang Province in both summer and winter.The spatial variationsmost of the temperature biases in HadGEM2-ES are relatively small relative to those in MIROC-based models.
In summer, the temperature is overestimated over eastern China, especially in northeastern China, northern Xinjiang, and the Tibetan Plateau, but underestimated over areas south of the Yangtze River in HadGEM2-ES (Fig. 3a).(Fig. 3c).In winter, however, the simulatedunderestimation of temperature in HadGEM2-ES is generally lower than that observed over Chinaexists at the national scale, with a regionally averaged cold bias of -3.041.79°C (Fig. 3d).in multi-model mean (Fig. 3f).Substantial cold biases occur over the Tarim Basin and the Tibetan Plateau, which are associated with regional topography.Most of the above biases are consistent among individual models, with the averaged model consistency of 76% over China in both summer and winter.
This indicates that although the injection of 5 Tg SO2 per year cannot returnleads to a surface cooling over China, the climatological temperature over China under RCP4.5 to in G4 is still higher than the present level, but can delay warming for several years.Considering that the feedback response timescale of diffusive ocean heat uptake in climate models is approximately ten years (Jarvis, 2011), simulationsthe simulation representing the last 40 years of injection (2030-2069) areis used to examine the temperature response to SAI over China, as done by Kravitz et al. (2013b) and Tilmes et al. (2013).
During this period, the warming trends over all of China in the G4 experimentamong models are 0.3621-0.4243°Cdecade -1 in summer and 0.3530-0.4859°Cdecade -1 in winter.As shown in Fig. 4, It can be seen that the warming trend differencesdifference between G4 and RCP4.5 areis small.This indicates that SAI in the G4 experiment has little impact on the warming, and this is expected because of the similar trend over China caused by GHG concentrations when the climate system reaches a relatively stable stateof radiative forcing variation in the two experiments during 2030-2069.The changes inregionally averaged temperature over China in G4 compared to RCP4.5 show that the strongest SAI-induced cooling occurs in HadGEM2-ES, with regional averages of is decreased by 0.24-0.96°C(0.64°C in the multi-model mean) in summer and 0.30-1.52°C(0.80°C in the multimodel mean) in winter.due to SAI forcing.Although the magnitude of SAI-induced temperature change varies across models and seasons, the cooling response is consistent among models over China.
The SAI-inducedwinter cooling in winter is also stronger than that inthe summer in level in all models.
Additionally, the result shows the strongest SAI-induced cooling occurs in HadGEM2-ES in bothMIROC-ESM-CHEM, with magnitudes of -0.61°C and -0.42°C, respectively.The cooling effect in winter (-0.56°C) is slightly weaker than that in summer (-0.60°C) in MIROC-ESMand winter.
The spatial patternspattern of the temperature differencesdifference between G4 and RCP4.5 over China areis illustrated in Fig. 5Figs and 7e).These increases are weak and insignificant.In summer, SAI-induced temperature changes are negative everywhere over China in all three models (Fig. 5a-c).In HadGEM2-ES, strong cooling occurs in the Yangtze-Huaihe River Basin and northern Xinjiang with magnitudes of -1.8 to -1.6°C (Fig. 5a).In MIROC-ESM and MIROC-ESM-CHEM, strong cooling with magnitudes of -1.1 to -0.8°C occurs in southeastern China and central Inner Mongolia, and the Tibetan Plateau, respectively (Fig. 5b-c).In winter, the cooling effect of SAI also occurs over all of China in HadGEM2-ES and MIROC-ESM-CHEM, with strong cooling over the Tibetan Plateau and northeastern China (up to -1.7°C), and northern China (up to -0.8°C), respectively (Fig. 5d-e).For MIROC-ESM, although SAI induces significant cooling over the southern Tibetan Plateau and northeastern China, its impact on winter temperature is weaker than -0.1°C and statistically insignificant in most of Central China.Moreover, the SAI induces slight warming over the source region of the Yellow River and the Sichuan Basin (Fig. 5f).The physical processes responsible for SAI-induced cooling or warming will be discussed in the subsequent sections.

Changes in surface energy components over ChinaDecomposition of SAI-induced temperature change
We decompose the SAI-induced change in surface air temperature over China by utilizing Eqs.
(1)-( 4).The regionally averaged value of each term is illustrated in Fig. 8.It can be seen that SAI decreases downward net surface radiation fluxes, leading to a surface cooling of 0.30-1.45°C in summer and 0.48-2.10°C in winter over China.These decreases are partly compensated by decreased nonradiative fluxes, especially the decreased LH.The contributions of SH, Q, and Res are relatively small (Fig. 8a).The decomposition of downward surface radiation shows the decreases in SW cs↓ and LW cs↓ in all models.The reduced LW cs↓ dominates the deficient downward net surface radiation and decreases the temperature with magnitudes of 0.38-1.33°C in summer and 0.25-1.38°C in winter.The reduced SW cs↓ also contributes to the surface cooling, with magnitudes of 0.04-0.33°C in summer and 0.13-0.41°C in winter.The winter decrease in SW cs↓ is stronger than the summer one in most models.
Besides, the inter-model differences in CRF and SAF changes are relatively substantial.The areaaveraged results illustrate that the changes in CRF and SAF have negative and positive contributions to the SAI-induced cooling over China in most models, respectively (Fig. 8b).
The spatial patterns of SAI-induced changes in key energy-related variables over China are illustrated in Fig. 9.Under SAI forcing, changes in atmospheric temperature and water vapor lead to a general decrease in the LW cs↓ .The SW cs↓ , primarily related to the solar radiation scattering effect by stratospheric sulfate aerosol particles, also exhibits a coherent reduction over China.The spatial pattern of temperature change over China is primarily determined by SW cl↓ and surface albedo changes.In summer, most models exhibit increases in cloud amount, especially over northwestern and central China.The resultant decreased SW cl↓ leads to strong cooling over these regions.Conversely, northeastern and southeastern China show increased SW cl↓ and relatively weak cooling (Fig. 9d).In MIROC-ESM-CHEM, the excessive SW cl↓ (up to 8 W m -2 ) offsets the clear-sky radiative effects and causes abnormal warming over most regions of eastern China (Fig. S1a).In summer, the surface albedo change due to SAI over China is relatively small.The increased surface albedo mainly occurs in the Tibetan Plateau, which contributes to local surface cooling (Fig. 9f).This may help to explain why the cloud effect is not a primary factor of temperature change over the Tibetan Plateau in summer.
In winter, a robust and coherent SAI-induced reduction in cloud cover is found over China (Fig. 9k).This reduction leads to a general increase in SW cl↓ , causing the weak cooling south of the Yangtze River valley.In other areas of China, however, the change in surface albedo is the primary factor affecting the spatial pattern of temperature response under SAI forcing.The increased surface albedo leads to strong cooling, especially over northwestern and central China.However, the decreased surface albedo is found over the upper reaches of the Yellow River and the middle and upper reaches of the Yangtze River in MIROC-ESM with magnitudes greater than 3%, which results in the abnormal winter warming mentioned above (Fig. S1d).Taken together, the increased summer cloud cover and winter surface albedo lead to strong cooling, while the decreased summer cloud cover and winter surface albedo result in weak cooling, or even warming for the certain subregions and models, for instance eastern China in MIROC-ESM-CHEM and the upper reaches of the Yellow River and the middle and upper reaches of the Yangtze River in MIROC-ESM.
We calculate the regional mean changes in surface air temperature, surface radiation, and surface turbulent heat fluxes due to SAI forcing over China (Fig. 6).A reduced net surface SW and an increased downward surface LH flux occur in all three models in both summer and winter, although their magnitudes vary.This indicates that the SAI-induced net surface SW deficit leads to surface cooling, while the deficient surface radiation is partly compensated by the increased downward LH flux over China.The decrease in clear-sky net surface SW (SW_CS), which is primarily related to the solar radiation scattering effect by stratospheric sulfate aerosol particles, is the main cause of the decreased SW.In contrast, the changes in the SW cloud radiative effect (SW_CRE) are positive and relatively small.Similarly, the decreased surface net LW contributes to surface cooling, excluding that occurred in HadGEM2-ES in winter.The regionally averaged changes in LW_CS and LW_CRE exhibit a distinctive difference between summer and winter.The decreased net surface LW is caused by the negative changes in LW_CS in summer, but by the negative changes in LW_CRE in winter.
The spatial patterns of SAI-induced changes in key energy-related variables over China are illustrated in Fig. 7 S2b).In addition, the spatial patterns of the net surface LW differences over China between G4 and RCP4.5 are consistent with those of the net surface SW differences in both summer and winter, but with opposite signs.The changes in LW are mainly dominated by those in LW_CS.One possible explanation is that the strong surface cooling caused by decreased net surface SW increases upward surface LW, which is enough to counteract the SAI-induced deficit of downward surface LW, leading to the positive net surface LW change.
In winter, a uniform SAI-induced reduction in cloud cover is found over China, with regional averages of -0.64% in HadGEM2-ES, -0.69% in MIROC-ESM, and -0.38% in MIROC-ESM-CHEM.This reduction leads to a general increase in the SW_CRE over China, causing the positive SW change south of the Yangtze River (Fig. 7k and S1k-S2k).In other areas of China, however, the changes in the net surface SW are closely related to SW_CS.The SAI-induced decrease in SW_CS leads to strong cooling over the Tibetan Plateau and northeastern China in HadGEM2-ES, northeastern China in MIROC-ESM, and northern China in MIROC-ESM-CHEM.Moreover, an increase in SW_CS is found over the source region of the Yellow River and the Sichuan Basin in MIROC-ESM with magnitudes greater than 6 W m -2 , leading to the abnormal winter warming mentioned above.
Altogether, the spatial pattern of SAI-induced temperature changes over China is mainly due to those in net surface SW.The deficit of net surface SW which leads to a strong surface cooling is mainly induced by the decreased SW_CRE in summer and the decreased SW_CS in winter.The exception is the strong surface cooling in summer over the Tibetan Plateau in MIROC-ESM-CHEM, which is related to the decreased SW_CS.In winter, abnormal warming is associated with a large positive SW_CS change in response to the SAI forcing.

Physical processes responsible for the SAI-induced SW change.temperature changes
Previous studies have illustrated that the SAI reduces the tropospheric temperature and atmospheric water vapor amount on a global scale (Kashimura et al., 2017;Visioni et al., 2018).In China, these reductions cause the decreased LW cs↓ , contributing to the surface cooling primarily.We further address the potential reasons for the SW cs↓ change by using the aforementioned decomposition method.The atmospheric reflection of solar radiation increases after sulfate aerosols injection.In our study, the effect of aerosols scattering on shortwave radiation is represented as SWSRM, which can be measured by the change in SAOD.As shown in Fig. 10, the latitudinal distributions of the calculated (used in HadGEM2-ES) and prescribed (used in BNU-ESM, CNRM-ESM1 and the MIROC-based models) SAOD changes caused by SAI in G4 display a coherent increase over China.The distribution in CanESM2 is not shown because it is a constant field according to the experimental design.The SAOD change in HadGEM2-ES is unavailable.Total aerosol optical depth is therefore considered as a reasonable alternative variable for SAOD (e.g., Bellouin et al., 2011).The national-scale increased SAOD results in a robust decrease in SWSRM (Figs. 11a, d), contributing to the surface cooling with magnitudes of 0.21-0.54°C in summer and 0.26-0.69°C in winter.Besides, the deficit in columnintegrated water vapor reduces the atmospheric absorption of solar radiation.The resultant increased SW (SWWV) counterbalance 37-81% and 11-48% of the reductions in SWSRM over China in summer and winter, respectively (Figs. 11b, e).This is the main reason why the SAI-induced winter cooling is severer than the summer level.Kashimura et al. (2017) pointed out that the net surface SW changes can at least explain a cooling of -1.1 to -0.2°C in response to the SAI forcing in the G4 experiment on a global scale.According to the above results, the spatial patterns of temperature differences over China between G4 and RCP4.5 are mainly determined by the net surface SW changes.The cloud radiative changes occurring over China have been discussed above.In this section, we further address other potential reasons for net surface SW changes by using the aforementioned decomposition method.Spatial patterns and regionally averaged values of the decomposition results over China are illustrated in Fig. 8 and S3-S4.
In response to the SAI forcing, changes in the solar radiation modification strength and surface albedo lead to a decrease in net surface SW (SWSRM and SWSA), while those in atmospheric water vapor and cloudiness lead to an increase in net surface SW (SWWV and SWC) over China in both summer and winter, although their magnitudes vary among models and seasons.Furthermore, the decrease in SWSRM is the largest contributor to the decreased net surface SW over China in all models, with magnitudes of -2.01 to -1.21 W m -2 in summer and -3.00 to -1.13 W m -2 in winter.
As discussed in Sect.4.2, the spatial patterns of summer and winter temperature changes over China are mainly determined by the SW cl↓ and surface albedo, respectively.Generally, the SAI-induced decrease in LH flux reduces the low cloud cover, resulting in the positive change in SW cl↓ (Figs.11c,     f).Through this process, the significantly decreased LH over northeastern and southeastern China causes the abnormal summer warming in MIROC-ESM-CHEM (Fig. S1c).However, in summer, the effect of LH is partly offset by the SAI-induced moisture convergence at the troposphere in most models.The resultant increased cloud cover enhances the surface cooling over northwestern and central China (Fig. 11h).The change in surface albedo is closely related to land surface conditions.
The SAI-induced cooling can be amplified by increased snow cover or sea ice (e.g., Schmidt et al., 2012).Considering surface albedo can be reasonably described as a linear function of snow cover fraction (Qu and Hall, 2007;Li et al., 2016), we further investigate the spatial pattern of changes in snow cover fraction, and find that matches with surface albedo over China (Figs. 11i,l; note that model data are not available for HadGEM2-ES).Under SAI forcing, the increased snow cover mainly occurs over the Tibetan Plateau in summer, and over northwestern and central China in winter.The enlarged snow cover fraction gives rise to SW decrease at the surface, which in turn has a positive feedback on surface cooling.Furthermore, the SAI-induced abnormal winter warming in MIROC-ESM is also associated with the decreased snow cover over the upper reaches of the Yellow River and the middle and upper reaches of the Yangtze River (Fig. S1e).
The spatial distributions of the SAI-induced SWSRM and SWWV changes show a general decrease and increase across China, respectively (Fig. 8 and S3-S4).The latitudinal distributions of the calculated (used in HadGEM2-ES) and prescribed (used in MIROC-based models) stratospheric AOD changes caused by SAI in the G4 experiment indicate a uniform increase in stratospheric AOD over China (Fig. 9).Note that the stratospheric AOD change in HadGEM2-ES is unavailable, and the tropospheric and stratospheric AOD change is therefore considered as a reasonable alternative (e.g., Bellouin et al., 2011).The increased AOD reflects the increased amount of stratospheric sulfate aerosol particles which leads to the decrease in the SWSRM.The results also show that the calculated value of the stratospheric AOD change in HadGEM2-ES is higher than the prescribed value in the MIROCbased models, which may be a primary cause of the strongest SAI-induced cooling over China in HadGEM2-ES in this study.Conversely, deficits in column-integrated water vapor over China caused by SAI occur in all models, with magnitudes of -2.28 to -0.88 kg m -2 in summer and -0.62 to -0.27 kg m -2 in winter.This reduction contributes to a decrease in atmospheric absorption of solar radiation, leading to the increase in SWWV (Fig. 10a-10f).
In addition to the SWSRM and SWWV, SWC and SWSA also play important roles in surface SW changes.The results indicate that changes in SWC and SWSA mainly determine the spatial pattern of net surface SW changes caused by SAI over China.The SWC changes, which are the same as the changes in SW_CRE, are discussed in Sect.4.2.The SWSA changes are closely related to land surface conditions.The SAI-induced increase in surface albedo in G4 leads to negative SWSA change over China (Fig. 10g-10l).In summer, the values of increased surface albedo are relatively small, with regional averages of 0.001 in HadGEM2-ES and 0.002 in MIROC-based models.A significantly increased surface albedo of up to 0.026 occurs in the Tibetan Plateau in MIROC-based models, which leads to decreased surface SW_CS and contributes to surface cooling.The regionally averaged increases in surface albedo range from 0.005 to 0.014 in winter.In addition, the aforementioned abnormal warming seen over the source region of the Yellow River and the Sichuan Basin in MIROC-ESM is also closely related to the decreased surface albedo (Fig. 10k).Considering surface albedo can be reasonably described as a linear function of snow cover fraction (e.g., Qu and Hall, 2007;Li et al., 2016), we further investigate the spatial pattern of differences in snow cover fraction in MIROC-ESM, and find that matches with surface albedo changes over China (Fig. S5; note that model data are not available for the other two models).It suggests that the SAI-induced surface albedo increase due to enlarged snow cover fraction gives rise to net surface SW decrease over China, which in turn has a positive feedback on surface cooling.

Conclusions and discussion
We analyze the surface air temperature response to SAI forcing over China based on the simulations fromof the G4 experiment and RCP4.5 scenario in three models (by using six GCMs (BNU-ESM, CanESM2, CNRM-ESM1, HadGEM2-ES, MIROC-ESM and MIROC-ESM-CHEM).
We also discuss the physical processes involved in the temperature response from a surface energy budget perspective.The main conclusions are summarized as follows.
(1) The threeAll selected models can well reproduce the present climatological surface air temperature over China in both summer and winter.The cooling effect caused byAlthough the SAI in the G4 experiment cannot returnleads to a surface cooling over China, the climatological temperature over China under RCP4.5 toin G4 is still higher than the present level but can delay warming for several years.Although the SAI-induced temperature differences between the G4 and RCP4.5 simulations are negative over China during.During the simulation period of 2030-2069, the cooling effect varies among models, regions and seasons.SAI leads to a national-scale cooling over China in all models.Regionally, the multi-model mean cooling is 0.64°C in summer and 0.80°C in winter, respectively.The SAI-induced temperature change varies among models, regions and seasons.In winter, the strong cooling in all three models, together with the abnormal warming in MIROC-ESM, is related to changes in SW_CS.Finally, note that equatorialEquatorial stratospheric sulfate aerosol geoengineering can induce global cooling through the transport of Brewer-Dobson circulation, and also leads to regional inequities in the temperature response due to the complicated processes of aerosol microphysics and stratospheric transport (Kravitz et al., 2019).This means that some areas will face more severe climatic disasters if this kind of geoengineering is implemented.To solve this issue, certain SAI experiments based on the regional injection method areat multiple locations have been proposed, such as the stratospheric aerosol geoengineering large ensemble project (GLENS) using CESM1(WACCM) (Tilmes et al., 2018).In addition, the uncertainty of the regional climate response to SAI is closely related to the reliability of the models (Irvine et al., 2016).It ishas been indicated that the CMIP6 GCMs perform better in simulating the temperature over China than their CMIP5 GCMscounterparts (Jiang et al., 2020).Therefore, the climate response to SAI geoengineering over China based on state-of-the-art GCM experiments merits further study.The dots indicate areas that are statistically significant at the 90% confidence level.whereat least twothirds of models share the same sign with the multi-model mean.
TOA SW  ), R is the fraction of reflection, A is the fraction of absorption during SW passing through the atmosphere, and α is the surface albedo.Considering that the four components of SW flux above (S, TOA SW  , SURF SW  and SURF SW  ) are directly available from model outputs, R, A, and α can be calculated from Eqs. ( as↓ represents downward surface shortwave radiation in all-sky conditions, SW cs↓ and LW cs↓ represent downward surface shortwave and longwave radiations in clear-sky conditions respectively, SW cl↓ and LW cl↓ represent downward shortwave and longwave radiative effects of clouds (all-sky radiations minus clear-sky radiations) respectively, and α represents surface albedo (the ratio of solar radiation reflected to the atmosphere at the surface).SAF is surface albedo feedback, and CRF is cloud radiative forcing.Under SAI forcing, both the changes in atmospheric reflection and atmospheric absorption affect the SW cs↓ .We assume that the clear-sky atmospheric reflection change is only affected by atmospheric water vapor amount, and the clear-sky atmospheric absorption change is only affected by the aerosol scattering effect.As detailed byKashimura et al. (2017), the change in SW cs↓ can be further decomposed as:It is noticeable that both R and A are affected by cloud cover.That is, both of them are all-sky values (R as and A as ).The clear-sky values of R and A (R cs and A cs ) are calculated using the clearseparate the effect of clouds.The effect of clouds, which is denoted as "cl", is calculated from the difference between all-sky and clear-sky values (e.g., cl as cs . 6 and 7.The multi-model results show a robust and coherent cooling in both summer and winter.Strong cooling with magnitudes greater than 0.8°C mainly occurs over high-latitude regions, including northwestern and central China.For the individual models, the SAI-induced temperature changes are negative and significant almost everywhere over China except for in MIROC-ESM and MIROC-ESM-CHEM.SAI leads to the temperature increases over the upper reaches of the Yellow River and the middle and upper reaches of the Yangtze River in MIROC-ESM in winter, and over northeastern and southeastern China in MIROC-ESM-CHEM in summer, respectively (Figs. 6f and S1-S2.In summer, the changes in net surface SW over China are closely related to those in SW_CRE.The regionally averaged changes in the cloud cover fraction over China show a consistent increase in all three models, with magnitudes of 0.10% in HadGEM2-ES, 0.04% in MIROC-ESM, and 0.06% in MIROC-ESM-CHEM.The increase in cloud cover, which mainly occurs in the Yangtze-Huaihe River Basin and northern Xinjiang in HadGEM2-ES and southeastern China and central Inner Mongolia in MIROC-ESM, induces a decrease in SW_CRE, leading to a strong cooling over these regions.Additionally, the deficit of the downward LH flux, especially over the Yangtze-Huaihe River Basin in HadGEM2-ES and central Inner Mongolia in MIROC-ESM, can increase cloud cover and amplify cooling although the regional mean changes in LH are positive.For the MIROC-based models, the large positive values of SW_CRE changes over the Tibetan Plateau are counteracted by decreased SW_CS, inducing a strong cooling in MIROC-ESM-CHEM (Fig. S1b and

( 2 )
The decomposition of temperature change based on the surface energy budget indicates that the SAI-induced surface cooling over China is dominated by the robust decrease in downward clearsky radiation fluxes (particularly in downward clear-sky longwave radiation flux), and associated with the changes in cloud effective forcing and surface albedo feedback.The shortwave radiative effect of clouds and the surface albedo feedback determine the spatial pattern of temperature change, which are somewhat model-dependent and display a level of regional and seasonal discrepancies.The regionally averaged surface radiation changes over China indicate that both the SAI-induced decreases in net surface SW and LW, except for the increased LW in winter in HadGEM2-ES, contribute to the surface cooling in all three models.In response to the SAI forcing, the spatial patterns of temperature changes over China are mainly induced by SW changes.In summer, the strong cooling in HadGEM2-ES and MIROC-ESM is mainly due to the decreased SW_CRE caused by the cloud cover decrease.The strong surface cooling over the Tibetan Plateau in MIROC-ESM-CHEM is related to the decreased SW_CS.

( 3 )
Under SAI forcing, the decreased downward clear-sky longwave radiation is mainly due to the decreased tropospheric temperature and water vapor amount, and the decreased downward clearsky shortwave radiation is mainly contributed by the aerosol scattering effect over China.The decreased latent heat flux generally reduces the cloud cover over China, but the change in summer cloud cover is closely associated with the anomalous tropospheric moisture flux convergence.The negative surface albedo feedback related to increased snow cover fraction also amplifies the surface cooling, especially over the Tibetan Plateau in summer, and over northwestern and central China in winter.The net surface SW decomposition shows that the increased SWSRM and SWSA and the decreased SWWV and SWC have positive and negative contributions to the decrease in net surface SW over China, respectively.Generally, SWSRM decreases and SWWV increases in both summer and winter, which are related to the increased stratospheric AOD and decreased column-integrated water vapor, respectively.The SWC and SWSA changes mainly determine the spatial patterns of SW changes due to SAI forcing.Moreover, both the strong summer cooling over the Tibetan Plateau in MIROC-ESM-CHEM and the abnormal winter warming in MIROC-ESM are related to the surface albedo changes.The results above are summarized schematically in Fig. 121.

Figure 1 .
Figure 1.Scatter plots of relationship between changes in surface air temperature (T) and surface temperature (Ts) over China due to SAI forcing during the period of 2030-2069 in (a) summer (JJA) and (b) winter (DJF), and CC is their correlation coefficient.Scatters and cross represent individual models and their mean,Schematic illustration representing the impacts of SAI geoengineering on net shortwave radiation flux at the surface.The SWSRM, SWWV, SWC, and SWSA represent the changes in shortwave radiation at the surface caused by those in solar radiation modification strength, amount of atmospheric water vapor, cloudiness, and surface albedo, respectively.

Figure 2 .
Figure 2. Taylor diagram of climatological seasonalsummer and winter temperatures over China between the historical simulations in selected models and observation during the present period of 1986-2005.Numbers represent individual models., and asterisks represent the multi-model mean.Red and blue represent summer and winter, respectively.The oblique dotted straight line shows the 99% confidence level determined from the two-tailed Student's t-test.

Figure 3 .
Figure 3. Spatial patterns of surface air temperature biasesclimatology (units: °C) over China between

Figure 4 .
Figure 4. Time series of regionally averaged surface air temperature (units: °C) over China in the G4 experiment (solid blue lines) and RCP4.5 scenario (solid red lines) in summer (JJA) and winter (DJF)..The values are obtained by subtracting the present climatology (mean of 1986-2005; represented in parentheses) in the historical experiment.Red and blue dashed lines represent the Theil-Senlinear trends of G4 and RCP4.5 simulations during the period of 2030-2069, respectively.The multi-model mean (MMM) is represented at the bottom, with the shading indicating one inter-model standard deviation.

Figure 109 .
Figure 109.Latitudinal distributions of the calculated (a, for HadGEM2-ES) and prescribed (b, for

Figure 1112 .
Figure 1112.Schematic diagram illustrating how the relatedrelevant physical processes impact the downward surface SWradiation changes over China in response to the SAI forcing in the G4 experiment.

Table 1 .
Main features and references of threeclimate models used in this study.