全球变化知识资源中心

The risks and benefits of solar geoengineering, or solar radiation management (SRM), depend on assumptions about its implementation. Claims that SRM will reduce precipitation, increase ocean acidification or deplete stratospheric ozone, or that it must be continued forever once started, are not inherent features of SRM; rather, they are features of common scenarios for its implementation. Most analyses assume, for example, that SRM would be used to stop the increase in global temperature or restore temperature to pre-industrial values. We argue that these are poor scenario choices on which to base policy-relevant judgements about SRM. As a basis for further analysis, we provide a scenario that is temporary in that its end point is zero SRM, is moderate in that it offsets only half of the growth in anthropogenic climate forcing and is responsive in that it recognizes that the amount of SRM will be adjusted in light of new information.

One cannot meaningfully evaluate solar geoengineering without a scenario for its implementation. It is now common, for example, to assert that more scientific research is needed to assess the balance between the risks and benefits of solar geoengineering, hereafter called solar radiation management (SRM). Yet the balance between risks and benefits depends at least as strongly on how SRM is deployed (for example on technology choice, timing and magnitude of the induced radiative forcing) as it depends on the climate's response to a specified SRM scenario.

Clear language is an essential tool for analysing this messy topic. We use SRM to denote a technology used to deliberately alter radiative forcing at sufficient scale to measurably alter the global climate. Any technology for producing radiative forcing will have a set of technology-specific impacts, such as ozone loss arising from the introduction of aerosol particles in the stratosphere. However the radiative forcing is produced, the efficacy of SRM is inherently limited by the fact that a change in solar radiative forcing cannot perfectly compensate for the radiative forcing caused by increasing greenhouse gases.

SRM has been variously framed as a substitute for cutting emissions (mitigation), as an emergency measure to be used if climate risks are higher than expected, or as a means to restoring surface temperatures to pre-industrial. Explicit or implicit, such scenarios shape any assessment of risk and efficacy of SRM.

Ocean acidification has been listed as a risk of SRM¹, yet acidification depends almost solely on cumulative CO₂ emissions and is unaffected by SRM. Ocean acidification is a risk of SRM only if SRM is used as a substitute for emissions mitigation; and in this case, the risk derives from the increase in emissions not from SRM.

Reduced precipitation is another frequently cited risk of SRM (see Supplementary Information for examples). It is true that if the SRM radiative forcing is large enough to offset all of the change in global mean temperature due to anthropogenic CO₂ — a common assumption — then precipitation will indeed be reduced in most locations². Simple physical arguments demonstrate that it takes a smaller SRM forcing to stop the rise in precipitation as CO₂ concentrations increase than is required to stop the rise in temperature³. Reduction in precipitation is, however, a product of the magnitude of SRM used in the scenario. If the SRM radiative forcing was adjusted to maintain global-average precipitation rates at their pre-industrial level then temperatures would be above pre-industrial. The claim that geoengineering will reduce average precipitation thus turns on the assumption that more SRM will be used than is required to stop the increase in precipitation caused by rising CO₂ concentrations.

As these examples illustrate, judgements about whether the use of SRM can be justified are determined by policy assumptions about how it will be used at least as strongly as they are determined by scientific analysis.

We articulate a scenario in sufficient detail to allow quantitative analysis of its physical and social implications, but we do not attempt to describe a political scenario that might result in this physical scenario being implemented. We do not claim that this scenario is likely or optimal, only that it is less suboptimal than the scenarios used most commonly. We adopt the central planner framing common in economic models that underlie much climate policy analysis and assume that decisions about implementation of SRM are made to maximize some measure of global welfare⁴. In practice, the nexus of decisions about SRM will involve nation states which are influenced by many factors, not least public and private transnational organizations, each of which have complex internal politics. Moreover, decisions about SRM take place in an environment in which decision makers face multiple issues and make decisions under substantial uncertainty. In this environment, the worst-case outcomes might include gross misuses of SRM or even war⁵.

Although we think it is unrealistic, we adopt the central planner framing for three reasons. First, because it is a common benchmark for climate policy analysis, it is a useful framework in which to compare SRM with other response options such as emissions mitigation and adaptation. Second, there is simply no tractable way to analyse the full decision problem, and our goal is not analysis but rather the construction of a scenario that is useful for further analysis including exploration of the political and institutional implications. Third, and finally, we hope that articulating an outcome that is closer to the social planner's optimum will aid the development of policy that might nudge the world towards a better outcome.

Our objective is to provide a scenario for implementation of SRM that is specific enough to be assessed and critiqued yet general enough to be used for a wide variety of science and policy analysis. We define the scenario in the next section while deferring the considerations that motivate our choice of scenario to the section following that. Next we explore a specific choice of scenario including technological details as a worked example. The final section provides a concluding summary.

Our scenario combines three elements: a specific method of altering solar forcing, an initial trajectory for SRM radiative forcing over time, and a plan for altering the trajectory based on new information. We aim to provide a scenario that is articulated in sufficient detail to allow quantitative evaluation of risk and efficacy.

Further, our scenario is chosen to meet the following criteria: (i) it is temporary in that the end point is zero SRM; (ii) it is moderate in that it does not offset all of the global mean temperature change due to increased greenhouse gases; and (iii) it is responsive in that it explicitly recognizes that the amount of SRM will be adjusted in light of new information. We elaborate the motivation behind each criterion in 'Guiding principles' below.

We link the amount of SRM to the amount of mitigation, in that slower growth in greenhouse gas forcing means a slower growth in SRM, but we do not make the converse linkage. We suggest that the risks and benefits of SRM be evaluated by comparing scenarios with and without SRM that use the same radiative forcing trajectory, although we recognize that the choice to use SRM may itself influence the amount of mitigation in one direction or the other. The scenario is defined as follows:

Radiative forcing trajectory. Beginning in 2020, adjust the global SRM radiative forcing so as to halve the rate of growth of net non-SRM anthropogenic radiative forcing. The top panel of Fig. 1 provides an example for a specific radiative forcing scenario.

The top panel shows the total radiative forcing for RCP4.5, and a radiative forcing profile in which the rate of change is halved starting in 2020; that is, for year k, RF_new(2020 + k) = RF_RCP4.5(2020 + k/2). The difference between these gives the suggested initial SRM profile in the second panel. The effect on global mean temperature as predicted by MAGICC (with a 3 °C climate sensitivity) is shown in the third panel, and the corresponding decadal rates of change in the final panel. Note that temperature and its rate of change would depend on climate sensitivity, but the amount of SRM would not. If rate-independent climate impacts increase superlinearly, then the benefits will be larger than is evident in the third panel. If impacts are quadratic in temperature, then impacts will be reduced by 20% in 2070 (roughly the time when SRM radiative forcing peaks) although the temperature increase ΔT is only reduced by 10%. (See ref. 16 for a climate damage function that depends on both the magnitude and rate of change of temperature.)

Three considerations shape our choice of scenario: moderation, responsiveness and impermanence.

Moderation (half measures). We define the benefits of SRM as the reduction in the magnitude or rate of climatic change due to anthropogenic greenhouse gas forcing; that is, the reduction in climate impacts. This definition is not trivial. Among other things, it ignores the fact that some regions or industries may benefit from anthropogenic climate change, and from their perspective a reduction in that climate change may therefore count as a harm rather than a benefit; and it ignores the extent to which the benefits (and harms) of climate change will be mediated by social, political, cultural and economic factors that themselves change as a consequence of social responses to climate change.

The impacts of climate change are primarily felt locally; that is, they depend on the local changes in variables such as temperature, precipitation and soil moisture. Analysis of the global utility of SRM therefore depends on how local benefits and harms are aggregated.

At one extreme, one can adopt the global optimal framing common in climate policy analysis. Under this assumption, the benefits of SRM first increase, then saturate and decline with increasing global radiative forcing. This holds true whatever weighting function is used to aggregate benefits across regions.

When the maximization of aggregate utility guides policy there is, usually, an implicit assumption that the winners will compensate the losers. The other extreme is Pareto's constraint, which states that a policy should be used to increase aggregate utility only so long as it makes no region worse off. For some choice of impact metric there will be regions that are worse off with any amount of SRM, so the Pareto-improving amount of SRM is zero. The same argument applies to mitigation: there are impact measures in which there are some winners as greenhouse gases increases, so the Pareto-improving amount of mitigation is zero. This extreme example serves as a warning against rigid application of Pareto's constraint. Almost any real-world public policy makes someone worse off. Rules that lie between global and Pareto optimality serve as better guides to policy than do either extreme. Analysis that demonstrates that the regional effectiveness of SRM is limited under Pareto's constraint should therefore be interpreted with caution^{12, 13}.

Impacts from climate change are typically assumed to increase faster than linearly with the magnitude of the change (for example quadratic in Nordhaus⁴, Weitzman¹⁴, Goes et al.¹⁵, cubic in Lempert et al.¹⁶). An example of the consequence of this assumption is shown in Fig. 2, where for illustration we have chosen a damage function that is quadratic in the local deviations of temperature and precipitation relative to a pre-industrial baseline. The specific choice of damage function is not critical to the argument, only the assumption that climate damages always increase faster than linearly, so that the 'benefits' of SRM (due to the intended reduction in climate changes) increase more slowly than linear, with the highest marginal benefit accruing initially.

Examples of the benefits and harms of SRM (left and right panels respectively) illustrating that benefits increase sublinearly and harms increase superlinearly. Whatever weighting is used to aggregate benefits and harms, the amount of SRM that maximizes the sum of benefits and harms will be less — perhaps much less — than the amount of SRM that maximizes benefits. The left panel shows a rough proxy for local climate damages, specifically the y-axis is the reduction in the sum of the mean of the quadratic deviation of temperature and precipitation across all climate model grid-points, where the deviation of each variable from its pre-industrial value is normalized by its interannual standard deviation. The climate model is HadCM3L (ref. 31) used with the methods and assumptions from ref. 8. The right panel shows chlorine activation as a function of surface area density (SAD) of sulphate aerosol computed using the AER model³² under mid-latitude lower-stratosphere conditions. Chlorine activation, a crucial determinant of ozone loss, is strongly determined by water vapour concentration. Anderson et al.¹⁷ provide a rationale for our choice of parameters. The secondary x-axis (top) shows an illustrative calculation of the corresponding solar reduction assuming that 0.5-μm radius sulphate aerosols were evenly dispersed over a 5-km height of altitude.

The above scenario and its justification are specific in terms of how to define the radiative forcing trajectory for SRM but not about how to produce it. To provide more context that might help in understanding this scenario, it is useful to consider some specifics about a particular way that it might be implemented, keeping in mind that this is only one possible approach, and there are other ways that would have different technology-specific impacts. Providing this level of detail on one possible approach serves to illustrate that: (i) the direct economic cost of initial SRM deployment would be so low that it is unlikely to play an important role in decisions by governments, and (ii) the technology development timeframe for initial deployment could be as short as a few years. This claim refers only to direct deployment costs and technological barriers; the costs of science and monitoring might exceed the cost of deployment for at least a decade, and the indirect benefits and harms of SRM are expected to be orders of magnitude larger.

Of the various approaches that have been suggested for SRM, the best understood is to introduce sulphate aerosol into the stratosphere. It is the only method that could be applied without substantial further technical development to generate global radiative forcing of a similar magnitude to greenhouse gas forcing.

The amount of sulphate aerosol required as a function of time depends on the forcing scenario and on the radiative forcing per unit of sulphate. For radiative forcing less than about 0.5 W m⁻² the radiative forcing efficacy is about 0.6–0.8 W m⁻² for an injection rate of one million tons of sulphur (MtS) per year for most proposed methods of introducing sulphate²⁵. So in the first decade of the scenario shown in Fig. 1 the rate at which the sulphur addition would increase — starting from zero in 2020 — would be 0.035 MtS yr⁻². That is, at the end of the first year the injection rate would be 0.035 MtS yr⁻¹ and after a decade it would be 0.35 MtS yr⁻¹.

Feedback control — responsiveness — could be used to ensure that the global radiative forcing increased as intended even if the efficacy per unit sulphur is uncertain. Measurement of radiative forcing from tropospheric aerosols is difficult because of their complex indirect effects on clouds, but stratospheric radiative forcing can probably be estimated with far greater accuracy. The aerosol distribution could, for example, be estimated from a combination of orbiting limb-sounders and lidars corroborated using in situ observations from which the radiative forcing could then be accurately estimated using optimal estimation methods. This approach rests on the fact that the main uncertainty in a priori estimates of radiative forcing for a given sulphate injection rate is in predicting the aerosol distribution, while prediction of radiative forcing given an aerosol distribution is far less uncertain.

Although delivery mechanisms have not been designed in detail, analysis suggests that the most cost-effective approach uses aircraft²⁶. Initially this might involve retrofitting business jets with off-the-shelf low-bypass ratio engines to allow them to fly at higher altitudes. Using McClellan and co-workers' analysis of re-engined G650 aircraft that include industry standard estimates of aircraft availability and flight rates, and assuming that the payload is liquid sulphur that is oxidized in situ, about two aircraft would be required in the first year, rising to 30 by 2040. The capital cost of purchasing and modifying these 30 aircraft would be roughly US$2.2 billion.

Deployment could begin with SO₂ but as the aerosol concentration increases, more of the added sulphate simply adds to the mass of existing aerosol, increasing aerosol size and so reducing the efficacy per unit sulphate²⁷. This problem can be avoided by direct release of H₂SO₄ from an aircraft as proposed by Pierce et al.²⁵. (Note that while the work of English et al.²⁸ appears to contradict this result, it simulates a process in which H₂SO₄ is perfectly mixed at the grid scale of the general circulation model, which does not — and would not be expected to — produce a result significantly different from the SO₂ oxidation case simulated by Heckendorn²⁷.)

If a decision were made to deploy SRM, we assume that efforts to develop new technologies would be pursued more actively so they might be available if problems with sulphate were worse than expected. This might include particles with less ozone impact, or particles with more efficient back-scattering (thus requiring fewer of them), or possibly space-based systems.

Critical to any plausible implementation scenario is monitoring of its effects so that either the amount or the implementation technology can be modified based on new information. As noted, it is much easier to detect the radiative forcing (or ozone chemistry impacts) than it is to detect the impact on regional climate variables such as temperature or precipitation because the former will have a much higher SNR. Any response that is too small to detect in the presence of natural variability should also not result in significant negative consequences, although any unusual weather extremes may be blamed on the SRM deployment nonetheless.

One approach that might help in evaluating the climate response due to SRM is to introduce some time-varying modulation of the SRM radiative forcing. The response due to SRM can then be estimated by looking for the correlated signal in any climate variable. It is easier to distinguish a time-varying response from background variability than it is to distinguish a secular change in SRM radiative forcing from similar changes in greenhouse gas concentrations. Even with modulation, it could take decades to be confident in attributing regional impacts to SRM¹⁸; however, modulation could allow earlier and more accurate detection of impacts on the chemistry and dynamics of the stratosphere where the signal-to-noise ratios will be much larger.

First and most simply, this scenario demonstrates that there may be value to temporary SRM. Humanity is not committed forever once SRM begins; rather, there is an implied commitment to a measured wind down rather than an abrupt termination.

We have explored SRM as a complement to mitigation in that we assume that SRM is used to reduce climate risks while mitigation proceeds. Here we compare this to other framings.

SRM is often considered as a substitute for mitigation. In the extreme case this means the use of SRM without any reduction of emissions. This could be effective in the short run but would be totally ineffective in the long run as greenhouse gas concentrations would rise without limit. More plausibly, SRM could be a partial substitute, although this entails risks due to the increased greenhouse gas concentrations. These risks are linked through the choice of policy although they are physically unrelated to SRM.

Alternately, SRM might be used only in case of a climate emergency²⁹. Our view is that if SRM is seriously contemplated (developed, governed and incorporated into climate policy) as an emergency measure, then it arguably makes more sense to begin some gradual and moderate SRM as a precursor. The reasons are primarily about providing time for learning. Reasons include: (i) starting early gives more time to learn about SRM effects, and how to do SRM better, as well as more time to learn about mitigation; (ii) starting at a small forcing amplitude provides a better environment for finding bad unknown-unknowns, as the consequences will be less severe at a small amplitude than at large; (iii) if there is a 'tipping point' beyond which climate impacts increase steeply, then the scenario described here would delay reaching it (assuming we are not already beyond it) and thus give more time to learn about it; and finally, (iv) moderate and gradual use of SRM provides a basis to develop governance mechanisms, whereas a 'climate emergency' might well be the worst circumstance for developing methods to govern a new technology like SRM.

The converse argument is that if SRM is intended only for emergencies then there is less chance it will be used. This is a preferred outcome if (i) the risks of SRM prove so large that even for partial temporary SRM they outweigh its benefits outside an emergency; or if (ii) socio-technical lock-in³⁰ is sufficiently strong that starting SRM amounts to a de facto commitment to use it at large scale. The central planner framing we have adopted here ignores the institutional factors that create strong lock-in.

Finally, we have shown that temporary SRM can be useful without use of carbon removal; but the technologies are complementary in that carbon removal can allow temporary SRM to limit both the rate and absolute magnitude of climate change.

The central message of this paper is not that the proposed scenario is likely or optimal, it is simply that analysis of SRM that is intended to inform policy should — at a minimum — be explicit about the implementation scenario that drives the analysis and about the way that conclusions are dependent on the scenario choice.

1. Robock, A. 20 reasons why geoengineering may be a bad idea. Bull. Atom. Sci. 64, 14–18 (2008).
   • Article
2. Bala, G., Duffy, P. B. & Taylor, K. E. Impact of geoengineering schemes on the global hydrological cycle. Proc. Natl Acad. Sci. USA 105, 7664–7669 (2008).