the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 15 Jan 2026
Robust assessment of Solar Radiation Modification risks and uncertainties must include shocks and societal feedbacks
Abstract.
- Preprint
(1320 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Review on egusphere-2026-28', Daniele Visioni, 19 Feb 2026
-
AC1: 'Response to Daniele Visioni', Benjamin Sanderson, 04 Mar 2026
Response to Review by Daniele Visioni
We thank Dr. Visioni for a detailed review that has strengthened this manuscript. We respond point-by-point below, quoting from revised text to illustrate the significant revisions to the paper.
________________
Point 1: Literature engagement and the "strawman" concern
"The authors, for some reason, have decided to completely ignore much of the conversation already happening in SRM research about scenarios, and essentially decided to build a strawman of it to criticize, instead of fairly reviewing already existing discussions, engaging with them and only at that point, of course, criticizing them."
This point on the discourse representation is well taken. We have substantially expanded the manuscript's engagement with this literature, and the introduction has been further restructured for clarity following coauthor input. The original submission did cite GeoMIP experiments (Kravitz et al., 2011), scenario frameworks (Visioni et al., 2024), and foundational works on SRM risks, but a fast-moving field of this breadth we agree demands more engagement, particularly with emulator-based scenario exploration and social science. The restructured introduction now follows a clearer four-part progression: (1) MIPs explore physical climate responses as their primary purpose; (2) policy-relevant futures require IAM/SSP frameworks whose enabling assumption is slow GHG timescales; (3) the SSP framework does not incorporate shocks, by design; (4) SRM breaks this timescale assumption, creating coupled governance-climate dynamics that current tools cannot resolve. Key passages:
"The climate modelling community has made substantial progress in exploring SRM's physical effects through coordinated experiments. [...] The primary purpose of such Model Intercomparison Projects is to foster understanding of the Earth system's response to SRM, providing essential and irreplaceable physical insights. Beyond this, they yield valuable policy-relevant information, but their design is oriented toward understanding physical climate responses, not toward exploring the plausibility space of coupled governance-climate futures."
We now include a paragraph on social science and governance:
"In parallel, a growing body of work has addressed the social and governance dimensions of SRM scenarios. Scenario classifications for politically relevant SRM futures have been proposed (Lockley et al., 2022), and research has examined the geopolitics of geoengineering governance (McLaren and Corry, 2021; Buck, 2022; Wiertz, 2015). Recent work has called for incorporating human perceptions of risk and behavioural dynamics into SRM scenario design (Beckage et al., 2025; Beckage et al., 2022), and for greater attention to social surprises in climate futures more broadly (Keys, 2023)."
The gap statement now reads:
"Despite this substantial and growing body of work across physical, social, and integrated modelling communities, a specific gap remains. The most critical, policy-relevant risks of SRM [...] while increasingly acknowledged and partially explored, are not yet represented as coupled governance-climate dynamics where physical outcomes and governance responses co-evolve on commensurate timescales. ESM experiments prescribe non-ideal scenarios exogenously and social science identifies the relevant dynamics qualitatively, but no existing framework integrates these into a coupled system."
And we acknowledge:
"This gap does not diminish the scientific value of existing work: the physical insights generated by idealised and non-ideal SRM experiments are essential foundations upon which any coupled framework must build."
A gap analysis in Section 4 now maps existing work against the SRMP typology:
"The bulk of current coordinated work, including GeoMIP experiments, multi-model intercomparisons, and their associated scenario design (MacMartin et al., 2022; Visioni et al., 2024; Kravitz et al., 2015) corresponds to SRMP-a, providing essential and irreplaceable physical understanding of idealised deployment. Emerging work on termination (Jones et al., 2013; Trisos et al., 2018), delayed deployment (Brody et al., 2024; Pfluger et al., 2024), and the emulator-based scenario exploration it enables (Farley et al., 2024; Farley et al., 2026) extends coverage toward SRMP-b. Recent unilateral deployment studies (Diao et al., 2025) begin to explore SRMP-c territory."
In total, we now cite 43 additional works relative to the original submission, including those prompted by the reviewer's suggestions. Organized by topic:
* Climate-society feedbacks: Beckage et al. (2018, 2022, 2025) on behavioural coupling and human perceptions in Earth system and SRM modelling; Farmer et al. (2015) on agent-based economic modelling of climate change; Müller-Hansen et al. (2017) on representing human decision-making in ESMs; Donges et al. (2020) and Moore et al (2022) on co-evolutionary modelling with endogenous human societies; Keys (2023) on social surprises; Tang et al. (2021)
* Model-predictive control and adaptive management: MacMartin (2019) on mission-driven research design for SRM; Kravitz et al. (2017), Richter et al. (2022), and Henry et al. (2023) on feedback-controller algorithms for multi-target SAI deployment (ARISE-SAI)
* Socioeconomic modelling of shocks: Hepburn et al. (2020), Kuzemko et al. (2020), Daniele et al. (2020) on COVID-19's socio-political dynamics; O'Neill et al. (2020) on shocks and wildcards in the SSP framework; Meinshausen et al. (2024), Rogelj (2013) on absence of climate impact feedbacks on socioeconomic development
* Moral hazard and risk-response feedbacks: McLaren (2016), Jebari et al. (2021), Andrews et al. (2022), Cherry et al. (2023), Abatayo et al. (2020), Acemoglu and Rafey (2018)
* Governance and conflict: McLaren and Corry (2021), Buck (2022), Wiertz (2015), Pezzoli et al. (2023), Castro et al. (2020), Lockley (2022)
* Global catastrophic risk: Halstead (2018), Tang and Kemp (2021), Kemp et al. (2022), Futerman and Beard (2023), Futerman et al. (2025)
* Philosophy of model robustness: Lloyd (2010)
* Decision making under deep uncertainty: Marchau et al. (2019), Lempert et al. (2024), Lempert and Prosnitz (2011), Bryant and Lempert (2010)
* Vulnerability and catastrophic shocks: Baum et al. (2013)
* SRM scenario design and physical modelling: MacMartin et al. (2022), Visioni et al. (2023a, 2023b, 2024), Brody et al. (2024), Pflüger et al. (2024), Quaglia et al. (2024), Diao et al. (2025), Laakso et al. (2016), Jones et al. (2013), Trisos et al. (2018), Farley et al. (2024, 2026), Samset et al. (2025), Gettelman et al. (2024), Estrada et al. (2026)
We acknowledge that the original submission missed some relevant references and thank the reviewers for identifying these gaps. However, we maintain that core issues raised in our paper, that of structuring an appropriate scenario space and modeling landscape to resolve climate-governance feedbacks and their effect on SRM deployment dynamics, is not covered by this literature.
________________
"This is just not true. MacMartin et al. (2022) simulated a broad range of scenarios, including termination, phase-out and multiple levels of cooling..."
We agree that the original sentence ("This leaves the most critical, policy-relevant risks of SRM [...] un-modeled by either ESM or IAM paradigms") was imprecise—several of these risks have been explored, though not yet as coupled dynamics. This sentence has been removed and replaced with the more precise formulation quoted above, distinguishing between risks that have been explored (termination, delayed deployment, volcanic eruptions during SAI, unilateral action, interactions with Earth system tipping elements) and the specific remaining gap of coupled governance-climate dynamics. We now acknowledge each of the specific works Dr. Visioni cites: MacMartin et al. (2022), Trisos et al. (2018), Quaglia et al. (2024), Laakso et al. (2016), Brody et al. (2024), Pfluger et al. (2024), Estrada et al. (2026), and Diao et al. (2025).
________________
"The discussion of how much and when to simulate non-ideal scenarios was often discussed in GeoMIP, which the authors do not acknowledge. For instance, see the review in Visioni et al. (2023a) (see Section 6.1.1, dedicated explicitly to this...) and in Visioni et al. (2024) (which the authors cite, but did not bother reading, because the reasoning behind the scenario choice were discussed with some attention, rather than being naive...)"
We agree that the word "naïve" was poorly chosen and have removed it along with other language that could be read as dismissive of deliberate methodological choices. The original conclusion passage ("This is why naïve assessment built on idealised assumptions poses such acute risks...") has been replaced with:
"This is why assessment based primarily on prescribed, idealised assumptions carries acute risks at this juncture. Existing physical climate studies and initial explorations of non-ideal scenarios provide essential foundations, but if the broader assessment of SRM does not extend to resolve the coupled governance-climate dynamics outlined here [...] important risk channels may remain inadequately characterised for decisions that could prove irreversible."
The phrase "systematically distorted evidence base" has also been removed, and the framing throughout now acknowledges the deliberate reasoning behind GeoMIP scenario choices rather than characterising them as uninformed.
However, the GeoMIP literature in many cases actually reinforces our core argument. Section 6.1.1 of Visioni et al. (2023a), "Is it necessary to simulate pathological scenarios in GeoMIP?", addresses precisely the question of whether GeoMIP should explore non-ideal governance-driven scenarios, arriving at the deliberate conclusion that it should not. For uncoordinated or "rogue actor" scenarios, the paper states: "we do not see a role for GeoMIP in these scenarios at this time." For termination experiments: "we presently see no need to conduct such investigations in numerous models." These are reasonable decisions given GeoMIP's mandate and resource constraints, but they confirm that the coupled governance-climate dynamics we identify in SRMP-c through SRMP-e are explicitly outside GeoMIP's self-defined purview.
Similarly, the scenario design criteria outlined in Section 6.2 (plausibility, policy relevance, scientific relevance, reproducibility) are oriented toward physical climate modelling concerns, and none address coupled governance-climate dynamics, governance feedbacks, or endogenous context shifts. The paper itself acknowledges this limitation, noting that "previous GeoMIP scenarios have been defined by a much smaller (and narrower in terms of expertise) community, mostly taking into account the modelers' needs" and calls for future scenarios to "carefully consider the need for an inclusive process that takes into account multiple lines of expertise across multiple fields (climate science, ecosystem sciences, social sciences)."
Our argument was never that GeoMIP's decisions were uninformed — we agree that they were clearly deliberate and well-reasoned within their scope. The SRMP framework identifies what must additionally be represented beyond that scope: the coupled, event-driven, governance-mediated dynamics that GeoMIP itself has chosen not to address. Far from contradicting our thesis, Section 6.1.1 corroborates it.
________________
"The authors also ignored large pieces of the literature in realms different from just climate science, curiously forgetting also the ones that offer various fair criticism of current scenarios [...] See for instance Wiertz et al. (2015), McLaren and Corry (2021), Buck (2022), Keys (2023), Beckage et al. (2025)..."
All five are now cited. We found Beckage et al. (2025) particularly valuable and cite both Beckage et al. (2025, 2022) as complementary efforts on coupled governance-climate dynamics.
________________
Point 2: Robustness, moral hazard, and the DOE comparison
"It would be interesting to know how the authors define 'robustness'. Personally, and also in the context of IPCC assessments, robustness is something I identify with perspectives such as Lloyd (2010, 2015) related to climate models' robustness: in her words robustness is the 'repeated production of the empirically successful model prediction or retrodiction against a background of independently-supported and varying model constructions'."
We now engage with Lloyd's definition directly. Revised Section 5:
"This distinction also clarifies our use of 'robustness.' In the philosophy of climate science, robustness refers to the 'repeated production of empirically successful model prediction or retrodiction against a background of independently-supported and varying model constructions' (Lloyd, 2010). Multi-model intercomparisons such as GeoMIP achieve this kind of robustness for physical climate responses, and their contributions are indispensable. We do not question this."
We then articulate three complementary dimensions:
"Our use of 'robust assessment' refers to a complementary concern on two levels. The first is process completeness: whether the assessment framework represents the classes of dynamics most relevant to the risk problem. For SRM, this means coupled governance-climate feedbacks (where governance responses and climate outcomes co-evolve on commensurate timescales) and stochastic shock responses (where discrete events such as governance failures or unilateral actions generate discontinuous trajectory changes). [...] The second is coverage adequacy: whether the space of futures explored is sufficient to characterise the risk landscape, given the processes represented. An assessment can be model-robust (well-validated across independent model constructions) yet process-incomplete (lacking the coupled feedback mechanisms that generate primary policy-relevant risks) and coverage-limited (unable to explore the futures those processes would produce). All three dimensions of robustness (model validation, process representation, and coverage) are necessary; none alone is sufficient. The SRMP framework addresses primarily the second and third dimensions, while acknowledging and building on the first."
The three-dimensional distinction (model validation, process representation, coverage) is central to the revised manuscript.
________________
"On the topic of moral hazard, which the authors often mention, it is important to note that they never specifically identify what they mean by it, nor do they acknowledge existing literature on the topic. [...] see for instance Andrews et al. (2022), Cherry et al. (2023); whereas if the authors wanted to partly support their assertions, they could point to Abatayo et al. (2020) and Acemoglu and Rafey (2018)."
To clarify: we do not have a "position" on whether moral hazard, or a suppression of mitigation, would materialise in response to SRM. Our framework is structurally agnostic on this question. The Mitigation Delay property (Section 4.1.2) is explicitly designed to take either sign: a high positive value represents mitigation deterrence, while a negative value represents "Spurring Ambition" where SRM awareness galvanizes stronger mitigation efforts. SRMP-a explicitly includes the possibility of accelerated mitigation under SRM. The framework requires only that this coupling be representable, not that it takes any particular value.
The revised manuscript foregrounds this bidirectionality:
"SRM deployment could reduce mitigation ambition ('mitigation deterrence' (McLaren, 2016) or 'risk-response feedback' (Jebari et al., 2021)), dynamically entrenching a high-emissions state; equally, SRM awareness could galvanize stronger mitigation efforts by making the climate problem more politically salient. The empirical evidence is genuinely mixed: experimental studies find that the anticipation of moral hazard can itself undermine mitigation cooperation (Andrews et al., 2022), that strategic uncertainty under SRM can lead to excessive cooling (Abatayo et al., 2020), and that the availability of geoengineering may not systematically reduce mitigation effort in experimental settings (Cherry et al., 2023). Theoretical work suggests that without credible commitment mechanisms, geoengineering availability distorts carbon taxation incentives (Acemoglu and Rafey, 2018). Our framework does not presuppose which direction this coupling takes. Rather, the fact that the mitigation-SRM coupling could take either sign, and that its sign may itself evolve over time, is precisely what makes it a 'Dynamic Context Loop' (Figure 2): the intervention can cause the world's socio-political trajectory to evolve in ways that static scenario assumptions cannot represent."
All four works are now cited. The framework provides the machinery to explore both moral hazard and mitigation acceleration, rather than assuming either outcome.
________________"Criticizing the scenarios used as 'unrealistic' to affirm that climate science assessments are wrong or biased was one of the central arguments of the now withdrawn 'A Critical Review of Impacts of Greenhouse Gas Emissions on the U.S. Climate' (Climate Working Group, 2025)."
We want to distance ourselves from this comparison unequivocally. Our argument is fundamentally different, and the revised manuscript makes this explicit. A new paragraph in Section 5:
"Our argument is not that existing SRM assessments based on idealised scenarios are invalid or biased; quite the opposite. The physical climate insights generated by coordinated programmes such as GeoMIP regarding climate sensitivity to SAI, termination dynamics, regional impact patterns, and inter-model uncertainties (Visioni et al., 2023a; Kravitz et al., 2015) are scientifically robust and policy-relevant on their own terms. The simplicity of well-designed idealised experiments is a scientific strength: it enables controlled diagnosis of physical uncertainties that more complex scenario designs would confound. Our argument is strictly additive: these essential physical foundations need to be complemented by frameworks that can additionally represent coupled governance-climate dynamics that current architectures were not designed to capture. The SRMP framework identifies what must be built on top of existing progress, not what should replace it."
The DOE document argued that existing assessments were biased because they used "unrealistic" scenarios. This is categorically not our position. We argue that existing assessments are valid and essential, and that additional representational capacity is needed for a complementary class of dynamics. The SRMP framework is positioned as extending the assessment landscape, not as correcting or replacing it.
________________The reviewer challenges our timescale argument, noting that we justify the adequacy of static scenarios for GHG assessments because "future emission trajectories are dominated by long-term developments including population dynamics, economic development and emission intensity of the economy" and "the climate system's response to GHG emissions is slowly emerging over multi-decadal horizons." The reviewer responds: "I don't think this is persuasive, especially since there is a lot of emerging science showing that, if one takes the broader view of short-lived forcers and not just CO2, climatic impacts arising from specific emission pathways can emerge quite soon (see Samset et al., 2025 or Gettelman et al., 2024 as examples). Does the emergence of these new understandings about rapid climate responses to mitigation mean that current climate change assessments do not meet 'minimum conditions for assessment' criteria? I don't think so, but it would flow from the authors' arguments."
A fair challenge. We do not claim that short-lived forcer dynamics invalidate GHG assessments: the timescale argument is about the dominant mode of the climate response, not the sole mode. Short-lived forcers operate on faster timescales but within a system where the dominant climate response (ocean heat uptake, ice dynamics) still integrates over decades. Samset et al. (2025) and Gettelman et al. (2024) demonstrate that aerosol reductions can produce rapid regional warming signals, but these signals emerge within a system whose long-term trajectory is governed by cumulative CO2 emissions and slow ocean dynamics. The fast aerosol response modulates the slow dominant response; it does not replace it.
The SRM case is qualitatively different in a specific structural sense: the primary climate intervention signal itself is as fast as the governance dynamics that control it. A termination event produces a climate response on the same timescale as the political decision that caused it, creating a tight bidirectional coupling between governance and climate that has no analogue in the GHG case. For short-lived GHG forcers, the fast climate signal is a consequence of emission decisions made for other reasons (economic restructuring, air quality regulation); for SRM, the fast climate signal is the intervention, and governance responses to that signal feed back directly into deployment decisions. This creates the coupled loop structure (Figure 2) that motivates our framework.
The revised text addresses this:
"An important nuance: not all anthropogenic climate forcing is slow. Rapid reductions in short-lived aerosol emissions, whether from shipping regulations or regional air quality policies, can produce detectable climate signals on sub-decadal timescales (Samset et al., 2025; Gettelman et al., 2024). However, these fast aerosol perturbations differ from intentional SRM in a structurally critical respect: they are side-effects of decisions made for other reasons (public health, environmental regulation), and the resulting climate signal does not feed back into the policy decision that caused it. No governance loop connects the warming produced by reduced shipping aerosols to a decision to reverse those reductions because of the warming. By contrast, SRM is defined by its intended climate effect: the entire purpose of the intervention is to produce a climate signal, and governance is directly responsive to the perceived success or failure of that signal. This intentionality creates a bidirectional coupling between SRM's rapid, sub-decadal climate effects and the governance dynamics that control deployment."
The emergence of fast aerosol climate signals does not imply that GHG assessments fail the "minimum conditions" test, because those signals lack the governance feedback loop that defines the SRM problem. The structural difference is not between slow and fast forcing, but between one-directional perturbations and bidirectionally coupled governance-climate systems.
________________
"With a large overlap in authorship list, ZECMIP is one of those (Jones et al., 2019), and now flat10MIP. I always found utterly puzzling the (widespread) criticisms of ZECMIP results about zero emission commitments because they're too idealized..."
We agree: the simplicity of ZECMIP's experimental design is a merit. "The simplicity of well-designed idealised experiments is a scientific strength." Our argument is not that idealised experiments need to be replaced by complex ones, but that an additional category of analysis, coupled governance-climate dynamics, is needed alongside them.
________________
Point 3: Practicality and the "so what" concern
"Section 4 concludes with the assertion that the authors are identifying 'what must be represented' to meet their threshold of a meaningful assessment. However, section 5 starts by saying that 'This paper does not prescribe a specific modeling solution'. So… what?"
The framework is diagnostic rather than prescriptive — deliberately so. We have strengthened the practical value in three ways.
First, a concrete gap analysis:
"To illustrate the diagnostic utility of this typology, we can map the coverage of existing SRM research against it. The bulk of current coordinated work [...] corresponds to SRMP-a, providing essential and irreplaceable physical understanding of idealised deployment. Emerging work on termination [...] extends coverage toward SRMP-b. Recent unilateral deployment studies (Diao et al., 2025) begin to explore SRMP-c territory. However, the coupled dynamics of SRMP-d (institutionalised inequality) and SRMP-e (mitigation substitution), in which governance feedbacks and context shifts are the primary risk mechanisms, remain largely unexplored in integrated frameworks."
This mapping makes visible which risk configurations are well-characterised, which are partially explored, and which represent critical blind spots. It also clarifies where physical climate modelling is the primary investigative tool (SRMP-a, b) and where coupled governance-climate approaches are additionally required (SRMP-c, d, e), directing different research communities toward the dimensions of the problem best suited to their methods.
Second, Section 5 has been substantially restructured around a new Figure 5, which replaces the original monolithic pipeline diagram with four panels showing typology-specific assessment architectures. Each panel identifies the tool classes, coupling mechanisms, and active feedback loops required for a distinct SRMP family: Panel A (SRMP-a/b) shows the existing pipeline with exogenous governance; Panel B (SRMP-c) introduces an N-agent strategic model with detection/attribution noise and a dynamic conflict state; Panel C (SRMP-d) maps regional equity perceptions feeding back into governance fragmentation; and Panel D (SRMP-e) models the self-reinforcing substitution cycle between SRM success and mitigation erosion. This directly addresses the commenter's concern that the paper must move beyond identifying the problem to showing how different modelling approaches could be applied to the developed architecture.
Updated Figure 5 — four-panel typology-specific suggested modeling framework is attached
Additionally - we add on the topic of constraint of research:
"The framework constrains rather than liberates. The seven bounded Driver Properties define the dimensions along which SRM risks vary, ruling out incoherent combinations and focusing analytical attention on specific risk mechanisms rather than unconstrained speculation. The five typologies provide a finite, parsimonious set of internally consistent governance logics that span the plausible SRM risk space. And the three-loop architecture provides concrete structural criteria (does an assessment represent stochastic events? endogenous context shifts? their coupling?) against which any proposed approach can be evaluated and existing experiments can be categorized."
On coherence:
"Critically, these typologies are not unconstrained: each represents an internally consistent governance logic where Fast Loop and Context Loop properties must cohere. A cooperative, globally coordinated deployment (SRMP-a) implies low conflict potential and low inequality potential by construction; a unilateral deployment (SRMP-c) cannot simultaneously assume cooperative governance."
________________"Ultimately then, as the authors say in the last phrase, the framework they themselves propose is not even an attempt at a solution. It's just pointing at a very complicated scheme on a board and saying 'see, this is just too complicated, let's just give up'."
We respectfully disagree with this characterisation. The SSP framework itself did not prescribe specific IAM architectures—it defined the structured space within which scenario development should proceed. Our framework functions analogously:
"In this sense, the SRMP framework functions analogously to how the SSP framework structures exploration of socioeconomic futures in scenarios: not by predicting outcomes, but by defining the wider socioeconomic and governance space within which the physical climate elements (for SRM, currently the best-characterised component) can be placed in context."
The conclusion of our paper is not "let's give up" but rather "here is what additionally needs to be built." We identify specific representational requirements (stochastic event generation, endogenous context evolution, their coupling to slow mitigation pathways) and map which risk configurations remain unaddressed. This is a research agenda, not a counsel of despair. We are actively developing modelling tools that operationalise the SRMP framework. Prior reviewers advised us to separate the conceptual framework from its implementation, and we followed that advice. The present paper establishes the diagnostic architecture; companion work on implementation is underway.
________________
"What the authors of this piece do is situate their proposal ambiguously between a simple classification of scenarios (but that was already done for instance in Lockley et al., 2022) and the proposal to actually integrate the fast-looped dynamics [...] But suggestions on how to do this exist: for instance, Beckage et al. (2025) proposed the inclusion of behavioral and cognitive processes in scenario exploration of SRM using social climate models (see Beckage et al., 2022)."
We now cite both Lockley et al. (2022) and Beckage et al. (2025, 2022). We view these as complementary rather than competing contributions. Lockley et al. (2022) provides a scenario classification; our framework adds the diagnostic architecture (the three-loop structure and seven Driver Properties) that specifies the mechanistic requirements for assessing those scenarios. Beckage et al. (2025) proposes a specific modelling methodology (social climate models incorporating behavioural dynamics); our framework identifies the broader structural requirements that any such methodology, including the Beckage approach, must satisfy. The Beckage proposal addresses part of the "Dynamic Context Loop" requirements we identify, particularly for behavioural feedbacks on mitigation ambition. Our framework additionally identifies requirements for stochastic shock generation, detection/attribution uncertainty, and geopolitical conflict dynamics that the social-climate-model approach does not currently address. We see these as collaborative, not competitive, research efforts.
________________"The authors need to make the case for why the specific realizations of chaos (the sub-optimal, failure-mode scenarios) they propose are a requirement for producing the generalized insights that an overarching assessment must return [...] Orderly, scientifically motivated experimental design can provide the greatest insights into the consequences of well-managed, idealized SRM as well as the failure modes in chaotic versions of SRM."
We agree that orderly experimental design is essential and that the GeoMIP approach of using idealised experiments to train emulators that can span the scenario space is a powerful and appropriate strategy. We do not advocate injecting chaos into ESM experiments. Our framework identifies representational requirements that apply to the full assessment pipeline—including emulators, reduced-complexity models, and analytical frameworks—not exclusively to ESM simulations. Meanwhile, we disagree with the implicit suggestion that because insights from an overarching assessment need to be sufficiently general, broad exploration of specific, contrasting futures would not be valuable. On the contrary, through a structured and broad exploration assessments can go beyond providing insights on the physical response of idealized SRM and also provide insights in the societal risks that emerge through the coupling of physical response and societal dynamics.
The emulator-based approach (Farley et al., 2024; 2026) is a compelling example: idealised GeoMIP experiments provide the physical basis, and emulators then enable rapid exploration of a broader scenario space. Our framework can inform what that broader space should include, by identifying which combinations of stochastic shocks and context shifts constitute the most policy-relevant risk configurations. In this sense, the SRMP typologies and existing emulator development are natural complements. The new Figure 5 makes this concrete: the emulator occupies the same structural role in all four panels, while what differs across typologies is the governance and feedback architecture surrounding it. The figure identifies the specific tool classes (game-theoretic models, equity-perception modules, mitigation-ambition modules) and coupling mechanisms required for each SRMP family, showing precisely how orderly experimental design feeds into progressively more complex assessment architectures.
________________
Concluding Remarks
The revised manuscript has been substantially restructured. The introduction now follows a four-part logical progression and engages extensively with the GeoMIP scenario design literature, emulator-based exploration, social science, and the GCR literature (43 additional citations). We have defined robustness explicitly via Lloyd (2010), distinguishing model-robustness from process completeness and coverage adequacy. The moral hazard discussion now presents mixed empirical evidence without presupposing a direction. The framework is positioned as strictly additive to existing work.
Beyond the literature engagement, the revision adds: a gap analysis mapping existing research against the SRMP typologies; coherence constraints on the typologies; a new Figure 5 with four typology-specific assessment architectures identifying tool classes and coupling mechanisms; substantive engagement with the DMDU literature (Marchau et al., 2019; Lempert et al., 2024); and a new subsection on deployment mechanism (SAI vs MCB) as a conditional modulator with Table 1.
________________
References cited in this response
* Abatayo, A. L. et al. (2020): Solar geoengineering may lead to excessive cooling and high strategic uncertainty. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.1916637117
* Acemoglu, D. and Rafey, W. (2018): Mirage on the horizon: Geoengineering and carbon taxation without commitment. NBER Working Paper No. 24411. https://doi.org/10.3386/w24411
* Andrews, T. M. et al. (2022): Anticipating moral hazard undermines climate mitigation in an experimental geoengineering game. Ecological Economics. https://doi.org/10.1016/j.ecolecon.2022.107421
* Beckage, B. et al. (2018): Linking models of human behaviour and climate alters projected climate change. Nature Climate Change. https://doi.org/10.1038/s41558-017-0031-7
* Beckage, B. et al. (2022): Incorporating human behaviour into Earth system modelling. Nature Human Behaviour. https://doi.org/10.1038/s41562-022-01478-5
* Baum, S. D. et al. (2013): Long-term trajectories of human civilization. Foresight. https://doi.org/10.1108/14636681311321121
* Bryant, B. P. and Lempert, R. J. (2010): Thinking inside the box: A participatory, computer-assisted approach to scenario discovery. Technological Forecasting and Social Change. https://doi.org/10.1016/j.techfore.2009.08.002
* Beckage, B. et al. (2025): Models and scenarios for solar radiation modification need to include human perceptions of risk. Environmental Research: Climate. https://doi.org/10.1088/2752-5295/addd42
* Brody, E. et al. (2024): Kicking the can down the road: understanding the effects of delaying the deployment of stratospheric aerosol injection. Environmental Research: Climate. https://doi.org/10.1088/2752-5295/ad53f3
* Buck, H. J. (2022): Environmental peacebuilding and solar geoengineering. Frontiers in Climate. https://doi.org/10.3389/fclim.2022.869774
* Castro, J. et al. (2020): A review of agent-based modeling of climate-energy policy. Wiley Interdisciplinary Reviews: Climate Change. https://doi.org/10.1002/wcc.647
* Cherry, T. L. et al. (2023): Climate cooperation in the shadow of solar geoengineering: An experimental investigation of the moral hazard conjecture. Environmental Politics. https://doi.org/10.1080/09644016.2022.2066285
* Daniele, G. et al. (2020): Wind of change? Experimental survey evidence on the COVID-19 shock and socio-political attitudes in Europe. CESifo Working Paper. https://doi.org/10.2139/ssrn.3671674
* Diao, C. et al. (2025): A model study exploring the decision loop between unilateral stratospheric aerosol injection scenario design and Earth system response. Earth's Future. https://doi.org/10.1029/2024EF005455
* Donges, J. F. et al. (2020): Earth system modelling with endogenous and dynamic human societies: the copan:CORE World-Earth modeling framework. Earth System Dynamics. https://doi.org/10.5194/esd-11-395-2020
* Estrada, F. et al. (2026): Economic assessment of SRM under socio-political and geophysical tipping dynamics. Environmental Research: Climate. https://doi.org/10.1088/2752-5295/ae33df
* Farley, A. et al. (2024): Emulating inconsistencies in stratospheric aerosol injection. Environmental Research: Climate. https://doi.org/10.1088/2752-5295/ad519c
* Farley, A. et al. (2026): A Climate Intervention Dynamical Emulator (CIDER). Geoscientific Model Development. https://doi.org/10.5194/egusphere-2025-1830
* Farmer, J. D. et al. (2015): A third wave in the economics of climate change. Environmental and Resource Economics. https://doi.org/10.1007/s10640-015-9965-2
* Futerman, G. et al. (2025): The interaction of solar radiation modification with Earth system tipping elements. Earth System Dynamics, 16, 939–978. https://doi.org/10.5194/esd-16-939-2025
* Futerman, L. and Beard, S. (2023): Geoengineering and global catastrophic risk. Futures. https://doi.org/10.1016/j.futures.2023.103206
* Gettelman, A. et al. (2024): Has reducing ship emissions brought forward global warming? Geophysical Research Letters. https://doi.org/10.1029/2024GL109077
* Halstead, J. (2018): Stratospheric aerosol injection research and existential risk. Futures. https://doi.org/10.1016/j.futures.2018.03.004
* Henry, M. et al. (2023): Comparison of UKESM1 and CESM2 simulations using the same multi-target stratospheric aerosol injection strategy. Atmospheric Chemistry and Physics. https://doi.org/10.5194/acp-23-13369-2023
* Hepburn, C. et al. (2020): Will COVID-19 fiscal recovery packages accelerate or retard progress on climate change? Oxford Review of Economic Policy. https://doi.org/10.1093/oxrep/graa015
* Jebari, J. et al. (2021): From moral hazard to risk-response feedback. Climate Risk Management. https://doi.org/10.1016/j.crm.2021.100324
* Jones, A. et al. (2013): The impact of abrupt suspension of solar radiation management (termination effect) in experiment G2 of the Geoengineering Model Intercomparison Project (GeoMIP). Journal of Geophysical Research: Atmospheres. https://doi.org/10.1002/jgrd.50762
* Jones, C. D. et al. (2019): The Zero Emissions Commitment Model Intercomparison Project (ZECMIP). Geoscientific Model Development. https://doi.org/10.5194/gmd-12-4375-2019
* Keys, P. W. (2023): The plot must thicken: a call for increased attention to social surprises in scenarios of climate futures. Environmental Research Letters. https://doi.org/10.1088/1748-9326/ace4e0
* Kemp, L. et al. (2022): Climate Endgame: Exploring catastrophic climate change scenarios. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.2108146119
* Kravitz, B. et al. (2011): The Geoengineering Model Intercomparison Project (GeoMIP). Atmospheric Science Letters. https://doi.org/10.1002/asl.316
* Kravitz, B. et al. (2015): The Geoengineering Model Intercomparison Project Phase 6 (GeoMIP6): Simulation design and preliminary results. Geoscientific Model Development. https://doi.org/10.5194/gmd-8-3379-2015
* Kravitz, B. et al. (2017): First simulations of designing stratospheric sulfate aerosol geoengineering to meet multiple simultaneous climate objectives. Journal of Geophysical Research: Atmospheres. https://doi.org/10.1002/2017JD026874
* Lempert, R. J. et al. (2024): Decision making under deep uncertainty for climate adaptation. Annual Review of Environment and Resources. https://doi.org/10.1146/annurev-environ-112321-095025
* Lempert, R. J. and Prosnitz, D. (2011): Governing geoengineering research: A political and technical vulnerability analysis of potential near-term options. RAND Corporation. https://doi.org/10.7249/TR846
* Kuzemko, C. et al. (2020): Covid-19 and the politics of sustainable energy transitions. Energy Research & Social Science. https://doi.org/10.1016/j.erss.2020.101685
* Laakso, A. et al. (2016): Radiative and climate impacts of a large volcanic eruption during stratospheric sulfur geoengineering. Atmospheric Chemistry and Physics. https://doi.org/10.5194/acp-16-305-2016
* Lloyd, E. A. (2010): Confirmation and robustness of climate models. Philosophy of Science. https://doi.org/10.1086/657427
* Lockley, A. et al. (2022): Politically relevant solar geoengineering scenarios. Socio-Environmental Systems Modelling. https://doi.org/10.18174/sesmo.18127
* MacMartin, D. G. and Kravitz, B. (2019): Mission-driven research for stratospheric aerosol geoengineering. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.1811022116
* MacMartin, D. G. et al. (2022): Scenarios for modeling solar radiation modification. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.2202230119
* McLaren, D. (2016): Mitigation deterrence and the "moral hazard" of solar radiation management. Earth's Future. https://doi.org/10.1002/2016EF000445
* McLaren, D. and Corry, O. (2021): Clash of geofutures and the remaking of planetary order: Faultlines underlying conflicts over geoengineering governance. Global Policy. https://doi.org/10.1111/1758-5899.12863
* Meinshausen, M. et al. (2024): A perspective on the next generation of Earth system model scenarios: towards Representative Emission Pathways (REPs). Geoscientific Model Development. https://doi.org/10.5194/gmd-17-4533-2024
* Marchau, V. A. W. J. et al. (2019): Decision Making under Deep Uncertainty: From Theory to Practice. Springer. https://doi.org/10.1007/978-3-030-05252-2
* Moore, Frances C., Katherine Lacasse, Katharine J. Mach, Yoon Ah Shin, Louis J. Gross, and Brian Beckage. "Determinants of emissions pathways in the coupled climate–social system." Nature 603, no. 7899 (2022): 103-111. https://doi.org/10.1038/s41586-022-04423-8
* Müller-Hansen, F. et al. (2017): Towards representing human behavior and decision making in Earth system models: an overview of techniques and approaches. Earth System Dynamics. https://doi.org/10.5194/esd-8-977-2017
* O'Neill, B. C. et al. (2020): Achievements and needs for the climate change scenario framework. Nature Climate Change. https://doi.org/10.1038/s41558-020-00952-0
* Pezzoli, A. et al. (2023): SRM on the table: The role of geoengineering for the stability and effectiveness of climate coalitions. Climatic Change. https://doi.org/10.1007/s10584-023-03604-2
* Pflüger, D. et al. (2024): Flawed emergency intervention: Slow ocean response to abrupt stratospheric aerosol injection. Geophysical Research Letters. https://doi.org/10.1029/2023GL106132
* Quaglia, I. et al. (2024): The potential of stratospheric aerosol injection to reduce the climatic risks of explosive volcanic eruptions. Geophysical Research Letters. https://doi.org/10.1029/2023GL107702
* Richter, J. H. et al. (2022): Assessing responses and impacts of solar climate intervention on the Earth system with stratospheric aerosol injection (ARISE-SAI). Geoscientific Model Development. https://doi.org/10.5194/gmd-15-8221-2022
* Rogelj, J., 2013. Uncertainties of low greenhouse gas emission scenarios - DISS. ETH NO. 21094 (Thesis). ETH Zurich, Zurich, Switzerland.
https://www.research-collection.ethz.ch/handle/20.500.11850/69139
* Samset, B. H. et al. (2025): East Asian aerosol cleanup has likely contributed to the recent acceleration in global warming. Communications Earth & Environment. https://doi.org/10.1038/s43247-025-02527-3
* Tang, A. et al. (2021): A fate worse than warming? Stratospheric aerosol injection and global catastrophic risk. Frontiers in Climate. https://doi.org/10.3389/fclim.2021.720312
* Trisos, C. H. et al. (2018): Potentially dangerous consequences for biodiversity of solar geoengineering implementation and termination. Nature Ecology & Evolution. https://doi.org/10.1038/s41559-017-0431-0
* Visioni, D. et al. (2023a): Opinion: The scientific and community-building roles of the Geoengineering Model Intercomparison Project (GeoMIP). Atmospheric Chemistry and Physics. https://doi.org/10.5194/acp-23-5149-2023
* Visioni, D. et al. (2023b): Climate response to off-equatorial stratospheric sulfur injections in three Earth system models. Atmospheric Chemistry and Physics. https://doi.org/10.5194/acp-23-663-2023
* Visioni, D. et al. (2024): G6-1.5K-SAI: A new Geoengineering Model Intercomparison Project (GeoMIP) experiment integrating recent advances in solar radiation modification studies. Geoscientific Model Development. https://doi.org/10.5194/gmd-17-2583-2024
* Wiertz, T. (2015): Visions of climate control: Solar radiation management in climate simulations. Science, Technology, & Human Values. https://doi.org/10.1177/0162243915606524
-
AC1: 'Response to Daniele Visioni', Benjamin Sanderson, 04 Mar 2026
-
CC1: 'Comment on egusphere-2026-28', Ben Kravitz, 19 Feb 2026
I’ll be honest; I struggled with this paper. I agree that a goal of better characterizing risks from SRM is useful. But if that’s the goal, I don’t think this paper meets the bar for either laying out the problem or proposing a solution. I unfortunately must recommend rejection. I will try not to reiterate many of the other reviewers’ points about strawman arguments, although I do share their concerns – the SRM community has done a lot of work that, in this paper, is uncredited.
Essentially, as far as I understand it, the authors propose a framework in which SRM can directly modify the underlying SSP via feedbacks. Good idea. But there are two major problems with this. First, it’s not a new idea – these are climate-society feedbacks, which the authors don’t seem to reference. Second, it’s not clear why this is a unique problem to SRM or why SRM is a useful stress test. The authors point out that the COVID-19 pandemic’s impact on climate was negligible. That was a pretty big shock to the system and highly disruptive throughout the world, as well as very difficult to predict or project. So can the authors provide a sense of why SRM would be a bigger or different shock, or provide some indication as to what modeling challenges it presents that COVID-19 didn’t? This paper presents that as a foregone conclusion, but I don’t think it’s obvious.
I think these issues come down to the authors’ conceptualization of risk and research to reduce risk. There are definitely risks that the current ESM and IAM frameworks don’t capture. But I don’t see why the proposed framework is the right answer. To me it looks so unconstrained (more on that below) that you could kind of get any answer you want, and that’s not a useful way to assess risk either. But again, I’m not sure why SRM specifically presents a problem for ESM/IAM coupling. This is a question that the IAM community has been wrestling with for at least 20 years (and there are numerous citations that I thought I would see about this), well before SRM was being seriously considered as a research topic by that community.
Ultimately it will be impossible to anticipate every single shock that could ever happen. The authors, as well as many other people researching SRM (some of which are cited) have indicated some options. But that doesn’t tell you much about risk. Risk is not a set of “what ifs”. It is a process in which you can use quantitative and qualitative tools to assign probabilities and consequences. For risks that are not reducible, there are techniques that people can use, such as forecasting, projection, and adaptive management. For some reason, these concepts go mostly unmentioned. It is a very important question, and one that the SRM community is specifically grappling with, as to which uncertainties can be reduced, which can be managed, and which must be tolerated.
Relatedly, there are vast swaths of literature that the authors are not referencing. Topics include climate-society feedbacks, model-predictive control, adaptive management, socioeconomic modeling (especially of shocks), risk, catastrophe theory, antecedents of conflict, governance mechanisms, etc. And, of course SRM itself (see the review by Visioni for specific citations). I don’t expect the authors to know everything. But I think proposing an idea requires understanding that many other people have done similar work. If the only goal is to provide a framework (which seems likely, as the authors say “This paper does not prescribe a specific modeling solution”) then I remain unconvinced that the framework is particularly novel, nor helpful for the main problems SRM is facing.
I have plenty of other comments on specific lines, but I don’t include them because I suspect that the revised paper would need to be so different as to make them moot. I think the authors could reformulate their arguments, in line with my comments and other reviewers’ comments, to make a useful contribution about how to quantify risks in SRM, which risks are indeed novel to SRM, and what modeling tools are necessary to address those risks. But unfortunately I don’t think they have succeeded at this task, and I struggle with understanding other purposes that their proposed framework serves.
Citation: https://doi.org/10.5194/egusphere-2026-28-CC1 -
AC3: 'Response to Ben Kravitz', Benjamin Sanderson, 04 Mar 2026
Thanks to Dr. Kravitz for a direct community comment that has prompted significant revisions. We address each point below, quoting from revised text to illustrate the changes in our paper following the reviews.
________________
Point 1: Novelty and climate-society feedbacks literature
"Essentially, as far as I understand it, the authors propose a framework in which SRM can directly modify the underlying SSP via feedbacks. Good idea. But there are two major problems with this. First, it's not a new idea, these are climate-society feedbacks, which the authors don't seem to reference."
We respectfully push back on the characterisation that we did not reference existing work on identifying climate-society feedbacks. The original paper cited O'Neill et al. (2020) on the need for the SSP framework to account for shocks and wildcards, as well as Meinshausen et al. (2024) on the absence of climate impact feedbacks on socio-economic development even in GHG-only regimes. Our paper was, from the outset, framed as a contribution to the climate-society feedback problem, not as a discovery of it.
That said, the original submission should have engaged more deeply with the specific literature the commenter has in mind, particularly model-predictive control, adaptive management, and behavioural modelling of climate-society feedbacks. The introduction has been substantially restructured following coauthor input as well as commenter feedback, and now follows a clearer four-part logical progression: (1) MIPs explore physical responses as their primary purpose; (2) policy-relevant futures require IAM/SSP frameworks whose enabling assumption is slow GHG timescales; (3) the SSP framework does not incorporate shocks, by design; (4) SRM breaks this timescale assumption, creating coupled governance-climate dynamics that current tools cannot resolve. Key passages:
"The climate modelling community has made substantial progress in exploring SRM's physical effects through coordinated experiments. [...] The primary purpose of such Model Intercomparison Projects is to foster understanding of the Earth system's response to SRM, providing essential and irreplaceable physical insights. Beyond this, they yield valuable policy-relevant information, but their design is oriented toward understanding physical climate responses, not toward exploring the plausibility space of coupled governance-climate futures."
"In parallel, a growing body of work has addressed the social and governance dimensions of SRM scenarios. Scenario classifications for politically relevant SRM futures have been proposed (Lockley, 2022), and research has examined the geopolitics of geoengineering governance (McLaren, 2021; Buck, 2022; Wiertz, 2015). Recent work has called for incorporating human perceptions of risk and behavioural dynamics into SRM scenario design (Beckage et al., 2025, 2022), and for greater attention to social surprises in climate futures more broadly (Keys et al., 2023)."
On the broader, non-SRM-specific challenge of climate-society feedbacks, we add the following:
"The challenge of endogenizing climate-society feedbacks is not unique to SRM: the broader integrated assessment and Earth system modelling communities have long recognized that treating socioeconomic development and climate as separable systems is a fundamental limitation, and various approaches have been proposed to close this loop, including behavioural coupling (Beckage et al., 2018), agent-based economic modelling (Farmer et al., 2015), frameworks for representing human decision-making in ESMs (Müller-Hansen et al., 2017), and modelling with endogenous human societies (Donges et al., 2020, Moore 2022)."
And, from the discussion of the SSP framework's treatment of shocks, we add:
"The persistence of this gap reflects not a lack of ambition but the difficulty of the problem, and the fact that the slow timescales of conventional mitigation have made the simplification of static socioeconomic contexts a tolerable approximation."
The gap statement now reads:
"Despite this substantial and growing body of work across physical, social, and integrated modelling communities, a specific gap remains. The most critical, policy-relevant risks of SRM [...] are not yet represented as coupled governance-climate dynamics where physical outcomes and governance responses co-evolve on commensurate timescales. ESM experiments prescribe non-ideal scenarios exogenously and social science identifies the relevant dynamics qualitatively, but no existing framework integrates these into a coupled system."
The revised manuscript now cites 43 additional works. We also now frame the timescale argument more precisely: the introduction explains how IAMs and the SSP framework emerged to explore policy-relevant futures under the enabling assumption that dominant GHG forcings are slow, and why this assumption breaks down for SRM. On novelty: we agree that climate-society feedbacks are well-established, and that the IAM community has long grappled with ESM/IAM coupling. Our contribution is the specific diagnostic architecture (the three-loop structure, seven Driver Properties, and five typologies) applied to SRM's particular challenge: the tight bidirectional coupling between fast governance dynamics and fast climate responses, which is structurally different from the slow feedbacks that IAM/ESM coupling typically addresses.
________________
Point 2: Why SRM is different from COVID-19 and other shocks
"Second, it's not clear why this is a unique problem to SRM or why SRM is a useful stress test. The authors point out that the COVID-19 pandemic's impact on climate was negligible. That was a pretty big shock to the system and highly disruptive throughout the world, as well as very difficult to predict or project. So can the authors provide a sense of why SRM would be a bigger or different shock, or provide some indication as to what modeling challenges it presents that COVID-19 didn't?"
We have substantially revised the COVID-19 discussion:
"If a pandemic, without any climate coupling, could trigger the kind of socio-political context shift that the SSP framework is not designed to represent, then SRM, where climate responses and governance dynamics are tightly coupled and mutually reinforcing, presents an even more compelling case for frameworks that can resolve these dynamics."
And, on the structural distinction between SRM and other fast forcings, we add:
"By contrast, SRM is defined by its intended climate effect: the entire purpose of the intervention is to produce a climate signal, and governance is directly responsive to the perceived success or failure of that signal. This intentionality creates a bidirectional coupling between SRM's rapid, sub-decadal climate effects and the governance dynamics that control deployment. The climate response becomes as fast as the political processes governing deployment, and those political processes respond directly to the perceived climate outcome."
The distinction is structural, not just quantitative. COVID-19 was a one-directional shock: it affected society but society's response to COVID-19 probably did not feed back through the physical climate to alter the shock itself. SRM shocks however could form a closed loop where the climate response to deployment is as fast as the political dynamics governing deployment, and those political dynamics respond directly to the perceived climate outcome. A termination event, for instance, produces a rapid climate signal that itself becomes the driver of further governance responses. This bidirectional coupling on commensurate timescales is what the SRMP framework's "Fast Loop" (Figure 2) is designed to represent, and it has no analogue in COVID-19 or other exogenous shocks.
The pandemic also strengthens our argument as a case study of context shifts, we add:
"The COVID-19 pandemic illustrates both the power and the limits of this design. As an acute global shock, it had very limited direct impact on the global climate (Forster et al., 2020), demonstrating that even severe, fast-timescale societal disruptions are effectively buffered by the slow physical climate response to GHG emissions: short-term shocks do not directly perturb the climate outcome. Yet COVID-19 was far from negligible as a socio-political event: it triggered massive fiscal reprioritisation that opened political space for both accelerated green investment and reactionary rollback (Hepburn et al., 2020), disrupted the politics of sustainable energy transitions and international cooperation (Kuzemko et al., 2020), and reshaped public trust in science and institutions (Daniele et al., 2020). It was precisely the kind of 'shock' that can shift a development pathway toward a different socioeconomic state (O'Neill et al., 2020), yet its consequences for the climate system emerge only gradually, through altered investment patterns and policy trajectories, not through direct climate coupling."
We now also address the short-lived aerosol forcer counterexample:
"Rapid reductions in short-lived aerosol emissions, whether from shipping regulations or regional air quality policies, can produce detectable climate signals on sub-decadal timescales (Samset et al., 2025; Gettelman et al., 2024). However, these fast aerosol perturbations differ from intentional SRM in a structurally critical respect: they are side-effects of decisions made for other reasons (public health, environmental regulation), and the resulting climate signal does not feed back into the policy decision that caused it. [...] By contrast, SRM is defined by its intended climate effect: the entire purpose of the intervention is to produce a climate signal, and governance is directly responsive to the perceived success or failure of that signal."
We agree some of these dynamics are not unique to SRM, and the paper now acknowledges this:
"However, the troublesome omission of fast, stochastic, human-system shocks is not unique to SRM. Geopolitical conflicts, global pandemics, disruptive technological breakthroughs, and financial crises are also fast-term dynamics that our current models average out."
SRM's distinction is not that it involves shocks (many systems do), but that it creates a specific structural coupling between fast climate responses and fast governance dynamics that makes the omission of these shocks a first-order problem for risk assessment rather than a second-order refinement.
________________
Point 3: The framework is unconstrained
"To me it looks so unconstrained (more on that below) that you could kind of get any answer you want, and that's not a useful way to assess risk either."
The revised manuscript now addresses this directly:
"The framework constrains rather than liberates. The seven bounded Driver Properties define the dimensions along which SRM risks vary, ruling out incoherent combinations and focusing analytical attention on specific risk mechanisms rather than unconstrained speculation. The five typologies provide a finite, parsimonious set of internally consistent governance logics that span the plausible SRM risk space. And the three-loop architecture provides concrete structural criteria (does an assessment represent stochastic events? endogenous context shifts? their coupling?) against which any proposed approach can be evaluated and existing experiments can be categorized."
On coherence, we add:
"Critically, these typologies are not unconstrained: each represents an internally consistent governance logic where Fast Loop and Context Loop properties must cohere. A cooperative, globally coordinated deployment (SRMP-a) implies low conflict potential and low inequality potential by construction; a unilateral deployment (SRMP-c) cannot simultaneously assume cooperative governance."
Put simply: you can't get any answer you want. The seven properties box you in, the five typologies rule out incoherent combinations, and the three-loop architecture gives concrete yes/no criteria for whether a given assessment captures the dynamics that matter. Furthermore - our new Figure 5 outlines exactly what model structure components are required to resolve each typology and class of risk.
________________
Point 4: Risk, probability, and adaptive management
"Risk is not a set of 'what ifs'. It is a process in which you can use quantitative and qualitative tools to assign probabilities and consequences. For risks that are not reducible, there are techniques that people can use, such as forecasting, projection, and adaptive management. For some reason, these concepts go mostly unmentioned."
Fair point. Revised Section 5:
"The SRMP framework is complementary to, not a replacement for, existing approaches to managing irreducible uncertainty in SRM research. Techniques such as adaptive management, model-predictive control, and mission-driven research design (MacMartin, 2019) are essential operational strategies for navigating SRM deployment under uncertainty. Indeed, several of these approaches, particularly robust decision-making and adaptive management, require a structured characterisation of the relevant risk space to function effectively (Marchau et al., 2019). The SRMP framework provides this: a systematic identification of the dimensions along which SRM risks vary and the governance logics that generate them, enabling more targeted application of these techniques."
The framework addresses a prior, structural question: what must be representable for these techniques to be applied meaningfully to the SRM problem? Adaptive management requires a characterisation of the space being managed; probabilistic risk assessment requires a delineation of the risk channels to be quantified. The SRMP typologies and Driver Properties provide this structured foundation.
On the feedback-control literature, the revised Section 5 now discusses this explicitly through the lens of Figure 5:
"A feedback controller modulates injection rates in response to climate output from an ESM or emulator, with governance treated as a fixed exogenous boundary condition and emissions provided by a conventional IAM. The Dynamic Context Loop is inactive: governance is assumed stable and cooperative, so there is no mechanism for shocks to alter the socio-political context. GeoMIP experiments, ARISE-SAI (Richter et al., 2022, Henry et al., 2023), and emulator-based scenario exploration (Farley et al., 2024, Farley et al., 2026) all operate within this architecture. It provides essential physical climate understanding but cannot represent the governance-mediated risks that define SRMP-c through SRMP-e."
Our contribution is to identify what additional dynamics must be representable when the "controller" is not an algorithm but a political system.
On the decision-theoretic gap, the revised manuscript now engages substantively with the Decision Making under Deep Uncertainty (DMDU) literature:
"The field of Decision Making under Deep Uncertainty (DMDU) offers a family of alternatives (Marchau et al., 2019): robust decision-making seeks strategies that perform acceptably across a wide range of plausible futures rather than optimally for any single forecast (Weaver et al., 2013); scenario discovery uses computational search to identify the combinations of uncertain parameters that most sharply distinguish policy success from failure (Bryant and Lempert, 2010); and dynamic adaptive policy pathways sequence near-term actions with pre-defined triggers for course correction. These approaches have already gained traction in IPCC assessment, particularly for adaptation and sea-level-rise planning (Lempert et al., 2024), and Lempert and Prosnitz (2011) demonstrated a preliminary application of robust decision-making to geoengineering governance."
Critically, all DMDU methods require a structured characterisation of the uncertainty space as input, and the SRMP typologies provide exactly this. The decision-theoretic gap is no longer left as an open question but connected to an established and active research programme.
________________
Point 5: Missing literature
"Relatedly, there are vast swaths of literature that the authors are not referencing. Topics include climate-society feedbacks, model-predictive control, adaptive management, socioeconomic modeling (especially of shocks), risk, catastrophe theory, antecedents of conflict, governance mechanisms, etc."
The revised manuscript now engages with substantially more literature across these domains, citing 42 additional works relative to the original submission. Organized by the topics the commenter identifies:
* Climate-society feedbacks: Beckage et al. (2018, 2022, 2025) on behavioural coupling and human perceptions in Earth system and SRM modelling; Farmer et al. (2015) on agent-based economic modelling of climate change; Müller-Hansen et al. (2017) on representing human decision-making in ESMs; Donges et al. (2020) on co-evolutionary World-Earth modelling with endogenous human societies; Keys (2023) on social surprises; Tang et al. (2021)
* Model-predictive control and adaptive management: MacMartin (2019) on mission-driven research design for SRM; Kravitz et al. (2017), Richter et al. (2022), and Henry et al. (2023) on feedback-controller algorithms for multi-target SAI deployment (ARISE-SAI)
* Socioeconomic modelling of shocks: Hepburn et al. (2020), Kuzemko et al. (2020), Daniele et al. (2020) on COVID-19's socio-political dynamics; O'Neill et al. (2020) on shocks and wildcards in the SSP framework; Meinshausen et al. (2024), Rogelj (2013) on absence of climate impact feedbacks on socioeconomic development
* Moral hazard and risk-response feedbacks: McLaren (2016), Jebari et al. (2021), Andrews et al. (2022), Cherry et al. (2023), Abatayo et al. (2020), Acemoglu and Rafey (2018)
* Governance and conflict: McLaren and Corry (2021), Buck (2022), Wiertz (2015), Pezzoli et al. (2023), Castro et al. (2020), Lockley (2022)
* Global catastrophic risk: Halstead (2018), Tang and Kemp (2021), Kemp et al. (2022), Futerman and Beard (2023), Futerman et al. (2025)
* Philosophy of model robustness: Lloyd (2010)
* Decision making under deep uncertainty: Marchau et al. (2019), Lempert et al. (2024), Lempert and Prosnitz (2011), Bryant and Lempert (2010)
* Vulnerability and catastrophic shocks: Baum et al. (2013)
* SRM scenario design and physical modelling: MacMartin et al. (2022), Visioni et al. (2023a, 2023b, 2024), Brody et al. (2024), Pflüger et al. (2024), Quaglia et al. (2024), Diao et al. (2025), Laakso et al. (2016), Jones et al. (2013), Trisos et al. (2018), Farley et al. (2024, 2026), Samset et al. (2025), Gettelman et al. (2024), Estrada et al. (2026)
We acknowledge that the original submission missed relevant references.
________________
Point 6: Novelty and utility of the framework
"If the only goal is to provide a framework (which seems likely, as the authors say 'This paper does not prescribe a specific modeling solution') then I remain unconvinced that the framework is particularly novel, nor helpful for the main problems SRM is facing."
Two additions address this.
First, a new Figure 5 (attached) replaces the original monolithic pipeline diagram with four panels showing typology-specific assessment architectures. Each panel identifies the tool classes, coupling mechanisms, and active feedback loops required for a distinct SRMP family: Panel A (SRMP-a/b) shows the existing pipeline with exogenous governance; Panel B (SRMP-c) introduces an N-agent strategic model with detection/attribution noise and a dynamic conflict state; Panel C (SRMP-d) maps regional equity perceptions feeding back into governance fragmentation; and Panel D (SRMP-e) models the self-reinforcing substitution cycle between SRM success and mitigation erosion. This is the kind of concrete, architecture-level specification the commenter was asking for.
Second, a concrete diagnostic gap analysis maps existing SRM research against the SRMP typologies:
"The bulk of current coordinated work, including GeoMIP experiments, multi-model intercomparisons, and their associated scenario design (MacMartin et al., 2022; Visioni et al., 2024; Kravitz et al., 2015), corresponds to SRMP-a, providing essential and irreplaceable physical understanding of idealised deployment. Emerging work on termination (Jones et al., 2013; Trisos et al., 2018), delayed deployment (Brody et al., 2024; Pfluger et al., 2024), and the emulator-based scenario exploration it enables (Farley et al., 2024; Farley et al., 2026) extends coverage toward SRMP-b. Recent unilateral deployment studies (Diao et al., 2025) begin to explore SRMP-c territory. However, the coupled dynamics of SRMP-d (institutionalised inequality) and SRMP-e (mitigation substitution), in which governance feedbacks and context shifts are the primary risk mechanisms, remain largely unexplored in integrated frameworks."
This mapping makes visible which risk configurations are well-characterised, which are partially explored, and which represent critical blind spots. It clarifies where physical climate modelling is the primary tool (SRMP-a, b) and where coupled governance-climate approaches are additionally required (SRMP-c, d, e).
Third, we position the framework as analogous to the SSP framework's role in structuring scenario development:
"In this sense, the SRMP framework functions analogously to how the SSP framework constrains socioeconomic scenario development: not by predicting outcomes, but by defining the wider socioeconomic and governance space within which the physical climate elements (for SRM, currently the best-characterised component) can be placed in context."
The SSP framework did not prescribe specific climate models or IAM architectures. Its contribution was to define the broader socioeconomic space (demographics, institutions, technology, policy) within which climate projections acquire meaning. Climate modelling is one element of that space, and arguably the most mature; the SSP framework's value lies in structuring everything around it. The SRMP framework functions analogously: the physical climate response to SRM is the best-understood component, but the governance dynamics, socioeconomic feedbacks, and stochastic disruptions that determine whether and how deployment occurs are what the framework structures. Different modelling approaches (including emulators, agent-based models, game-theoretic frameworks, and social-climate models such as those proposed by Beckage et al., 2025) can then be evaluated against these wider representational requirements.
We are actively developing modelling tools that operationalise the SRMP framework. Prior reviewers advised us to separate the conceptual framework from its implementation, and we followed that advice. The present paper establishes the diagnostic architecture; companion work on implementation is underway.
________________
Point 7: The "which uncertainties can be reduced, managed, or tolerated" question
"It is a very important question, and one that the SRM community is specifically grappling with, as to which uncertainties can be reduced, which can be managed, and which must be tolerated."
Agreed — this is complementary to our contribution. The SRMP framework organises exactly this question:
* Reducible uncertainties (physical climate response, aerosol microphysics) are primarily addressed through SRMP-a and SRMP-b class research, using idealised experiments and multi-model intercomparisons. Existing approaches are well-suited to this task.
* Manageable uncertainties (operational deployment challenges, detection and attribution) require the adaptive management, model-predictive control, and mission-driven research design approaches that the commenter rightly flags. The Fast Loop Driver Properties (Interruption Risk, Deployment Regionality, Operational Inefficacy, Detection/Attribution Uncertainty) provide the structured dimensions against which these management strategies can be targeted.
* Uncertainties that must be tolerated (governance failures, geopolitical conflict, mitigation coupling) are precisely the coupled dynamics that SRMP-c through SRMP-e address. For these, the framework's contribution is to identify and structure them so that robust decision-making approaches can at least characterise the risk landscape, even where prediction is impossible. For these uncertainties, scenarios and storylines are useful, because they allow for the construction of 'what-if' experimental designs which can simulate the quantifiable damages conditional on the unquantifiable risk of societal collapse.
The SRMP framework does not resolve which uncertainties fall into which category. It provides the structured risk space within which that triage can be performed.
________________
Point 8: Reformulation suggestion
"I think the authors could reformulate their arguments, in line with my comments and other reviewers' comments, to make a useful contribution about how to quantify risks in SRM, which risks are indeed novel to SRM, and what modeling tools are necessary to address those risks."
Thanks for this suggestion. The revised manuscript addresses each of these three goals directly:
1. How to quantify risks in SRM: The gap analysis in Section 4 does what the commenter asks for — it maps existing work against the SRMP typologies and shows where the blind spots are. SRMP-a and -b are well covered by existing physical modelling. SRMP-c is starting to get attention (Diao et al., 2025). SRMP-d and -e (the coupled governance-climate dynamics) are essentially unaddressed in current literature. The framework also connects to the DMDU literature (Marchau et al., 2019; Lempert et al., 2024; Bryant and Lempert, 2010), so the evaluation methodology is no longer an open question.
2. Which risks are novel to SRM: Not shocks per se — those are general. What is novel is the tight bidirectional loop: the climate signal is the intervention, governance responds to it in real time, and that response changes the climate signal. COVID-19 was a massive shock but one-directional — society's response did not feed back through climate to alter the shock. Short-lived aerosol forcers produce fast climate signals but are side-effects, with no governance loop closing back through the climate. SRM closes the loop.
3. What modelling tools are necessary: The new Figure 5 answers this concretely. Four panels, four architectures, each with specific tool classes and coupling mechanisms: feedback controllers for SRMP-a/b (the existing pipeline), N-agent strategic models for SRMP-c (conflict), equity-perception modules for SRMP-d (free driver), and endogenous mitigation-ambition models for SRMP-e (moral hazard). The text engages directly with the commenter's own work on feedback-control methods (Kravitz et al. 2017, Richter et al. 2022, Henry et al. 2023), as well as adaptive management (MacMartin 2019), game theory (Pezzoli et al. 2023), and social-climate models (Beckage et al. 2022, 2025). The relationship is straightforward: physical climate modelling is the best-understood piece; the SRMP framework identifies what needs to be built around it — the same way the SSP framework defines the socioeconomic space around climate projections, not the projections themselves.
Updated Figure 5 — four-panel typology-specific assessment architectures attached
________________
Concluding Remarks
The revision engages extensively with the climate-society feedback, adaptive management, feedback-control, and DMDU literatures (42 additional citations). The COVID-19 and short-lived aerosol discussions now sharply distinguish the bidirectional governance-climate coupling unique to SRM from exogenous shocks and one-directional fast forcers. The framework is explicitly constrained through internal coherence requirements, bounded Driver Properties, and finite typologies, and positioned as complementary to adaptive management, model-predictive control, and ARISE-SAI feedback-controller methods.Structurally, the revision adds: a gap analysis mapping existing research against the SRMP typologies; a new Figure 5 with four typology-specific assessment architectures; engagement with Lloyd (2010) distinguishing model-robustness from process completeness and coverage adequacy; connection to DMDU evaluation methods (Marchau et al., 2019; Lempert et al., 2024); and a new subsection on deployment mechanism (SAI vs MCB) as a conditional modulator with Table 1.
________________
References cited in this response
* Abatayo, A. L. et al. (2020): Solar geoengineering may lead to excessive cooling and high strategic uncertainty. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.1916637117
* Acemoglu, D. and Rafey, W. (2018): Mirage on the horizon: Geoengineering and carbon taxation without commitment. NBER Working Paper No. 24411. https://doi.org/10.3386/w24411
* Andrews, T. M. et al. (2022): Anticipating moral hazard undermines climate mitigation in an experimental geoengineering game. Ecological Economics. https://doi.org/10.1016/j.ecolecon.2022.107421
* Beckage, B. et al. (2018): Linking models of human behaviour and climate alters projected climate change. Nature Climate Change. https://doi.org/10.1038/s41558-017-0031-7
* Baum, S. D. et al. (2013): Long-term trajectories of human civilization. Foresight. https://doi.org/10.1108/14636681311321121
* Bryant, B. P. and Lempert, R. J. (2010): Thinking inside the box: A participatory, computer-assisted approach to scenario discovery. Technological Forecasting and Social Change. https://doi.org/10.1016/j.techfore.2009.08.002
* Beckage, B. et al. (2022): Incorporating human behaviour into Earth system modelling. Nature Human Behaviour. https://doi.org/10.1038/s41562-022-01478-5
* Beckage, B. et al. (2025): Models and scenarios for solar radiation modification need to include human perceptions of risk. Environmental Research: Climate. https://doi.org/10.1088/2752-5295/addd42
* Brody, E. et al. (2024): Kicking the can down the road: understanding the effects of delaying the deployment of stratospheric aerosol injection. Environmental Research: Climate. https://doi.org/10.1088/2752-5295/ad53f3
* Buck, H. J. (2022): Environmental peacebuilding and solar geoengineering. Frontiers in Climate. https://doi.org/10.3389/fclim.2022.869774
* Castro, J. et al. (2020): A review of agent-based modeling of climate-energy policy. Wiley Interdisciplinary Reviews: Climate Change. https://doi.org/10.1002/wcc.647
* Cherry, T. L. et al. (2023): Climate cooperation in the shadow of solar geoengineering: An experimental investigation of the moral hazard conjecture. Environmental Politics. https://doi.org/10.1080/09644016.2022.2066285
* Daniele, G. et al. (2020): Wind of change? Experimental survey evidence on the COVID-19 shock and socio-political attitudes in Europe. CESifo Working Paper. https://doi.org/10.2139/ssrn.3671674
* Diao, C. et al. (2025): A model study exploring the decision loop between unilateral stratospheric aerosol injection scenario design and Earth system response. Earth's Future. https://doi.org/10.1029/2024EF005455
* Donges, J. F. et al. (2020): Earth system modelling with endogenous and dynamic human societies: the copan:CORE World-Earth modeling framework. Earth System Dynamics. https://doi.org/10.5194/esd-11-395-2020
* Estrada, F. et al. (2026): Economic assessment of SRM under socio-political and geophysical tipping dynamics. Environmental Research: Climate. https://doi.org/10.1088/2752-5295/ae33df
* Farley, A. et al. (2024): Emulating inconsistencies in stratospheric aerosol injection. Environmental Research: Climate. https://doi.org/10.1088/2752-5295/ad519c
* Farley, A. et al. (2026): A Climate Intervention Dynamical Emulator (CIDER). Geoscientific Model Development. https://doi.org/10.5194/egusphere-2025-1830
* Farmer, J. D. et al. (2015): A third wave in the economics of climate change. Environmental and Resource Economics. https://doi.org/10.1007/s10640-015-9965-2
* Futerman, G. et al. (2025): The interaction of solar radiation modification with Earth system tipping elements. Earth System Dynamics, 16, 939–978. https://doi.org/10.5194/esd-16-939-2025
* Futerman, L. and Beard, S. (2023): Geoengineering and global catastrophic risk. Futures. https://doi.org/10.1016/j.futures.2023.103206
* Forster, P. M. et al. (2020): Current and future global climate impacts resulting from COVID-19. Nature Climate Change. https://doi.org/10.1038/s41558-020-0883-0
* Gettelman, A. et al. (2024): Has reducing ship emissions brought forward global warming? Geophysical Research Letters. https://doi.org/10.1029/2024GL109077
* Halstead, J. (2018): Stratospheric aerosol injection research and existential risk. Futures. https://doi.org/10.1016/j.futures.2018.03.004
* Henry, M. et al. (2023): Comparison of UKESM1 and CESM2 simulations using the same multi-target stratospheric aerosol injection strategy. Atmospheric Chemistry and Physics. https://doi.org/10.5194/acp-23-13369-2023
* Hepburn, C. et al. (2020): Will COVID-19 fiscal recovery packages accelerate or retard progress on climate change? Oxford Review of Economic Policy. https://doi.org/10.1093/oxrep/graa015
* Jebari, J. et al. (2021): From moral hazard to risk-response feedback. Climate Risk Management. https://doi.org/10.1016/j.crm.2021.100324
* Jones, A. et al. (2013): The impact of abrupt suspension of solar radiation management (termination effect) in experiment G2 of the Geoengineering Model Intercomparison Project (GeoMIP). Journal of Geophysical Research: Atmospheres. https://doi.org/10.1002/jgrd.50762
* Keys, P. W. (2023): The plot must thicken: a call for increased attention to social surprises in scenarios of climate futures. Environmental Research Letters. https://doi.org/10.1088/1748-9326/ace4e0
* Kemp, L. et al. (2022): Climate Endgame: Exploring catastrophic climate change scenarios. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.2108146119
* Kravitz, B. et al. (2015): The Geoengineering Model Intercomparison Project Phase 6 (GeoMIP6): Simulation design and preliminary results. Geoscientific Model Development. https://doi.org/10.5194/gmd-8-3379-2015
* Kravitz, B. et al. (2017): First simulations of designing stratospheric sulfate aerosol geoengineering to meet multiple simultaneous climate objectives. Journal of Geophysical Research: Atmospheres. https://doi.org/10.1002/2017JD026874
* Lempert, R. J. et al. (2024): Decision making under deep uncertainty for climate adaptation. Annual Review of Environment and Resources. https://doi.org/10.1146/annurev-environ-112321-095025
* Lempert, R. J. and Prosnitz, D. (2011): Governing geoengineering research: A political and technical vulnerability analysis of potential near-term options. RAND Corporation. https://doi.org/10.7249/TR846
* Kuzemko, C. et al. (2020): Covid-19 and the politics of sustainable energy transitions. Energy Research & Social Science. https://doi.org/10.1016/j.erss.2020.101685
* Laakso, A. et al. (2016): Radiative and climate impacts of a large volcanic eruption during stratospheric sulfur geoengineering. Atmospheric Chemistry and Physics. https://doi.org/10.5194/acp-16-305-2016
* Lloyd, E. A. (2010): Confirmation and robustness of climate models. Philosophy of Science. https://doi.org/10.1086/657427
* Lockley, A. et al. (2022): Politically relevant solar geoengineering scenarios. Socio-Environmental Systems Modelling. https://doi.org/10.18174/sesmo.18127
* MacMartin, D. G. and Kravitz, B. (2019): Mission-driven research for stratospheric aerosol geoengineering. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.1811022116
* MacMartin, D. G. et al. (2022): Scenarios for modeling solar radiation modification. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.2202230119
* McLaren, D. (2016): Mitigation deterrence and the "moral hazard" of solar radiation management. Earth's Future. https://doi.org/10.1002/2016EF000445
* McLaren, D. and Corry, O. (2021): Clash of geofutures and the remaking of planetary order: Faultlines underlying conflicts over geoengineering governance. Global Policy. https://doi.org/10.1111/1758-5899.12863
* Meinshausen, M. et al. (2024): A perspective on the next generation of Earth system model scenarios: towards Representative Emission Pathways (REPs). Geoscientific Model Development. https://doi.org/10.5194/gmd-17-4533-2024
* Marchau, V. A. W. J. et al. (2019): Decision Making under Deep Uncertainty: From Theory to Practice. Springer. https://doi.org/10.1007/978-3-030-05252-2
* Müller-Hansen, F. et al. (2017): Towards representing human behavior and decision making in Earth system models: an overview of techniques and approaches. Earth System Dynamics. https://doi.org/10.5194/esd-8-977-2017
* Moore, Frances C., Katherine Lacasse, Katharine J. Mach, Yoon Ah Shin, Louis J. Gross, and Brian Beckage. "Determinants of emissions pathways in the coupled climate–social system." Nature 603, no. 7899 (2022): 103-111. https://doi.org/10.1038/s41586-022-04423-8
* O'Neill, B. C. et al. (2020): Achievements and needs for the climate change scenario framework. Nature Climate Change. https://doi.org/10.1038/s41558-020-00952-0
* Pezzoli, A. et al. (2023): SRM on the table: The role of geoengineering for the stability and effectiveness of climate coalitions. Climatic Change. https://doi.org/10.1007/s10584-023-03604-2
* Pflüger, D. et al. (2024): Flawed emergency intervention: Slow ocean response to abrupt stratospheric aerosol injection. Geophysical Research Letters. https://doi.org/10.1029/2023GL106132
* Rogelj, J., 2013. Uncertainties of low greenhouse gas emission scenarios - DISS. ETH NO. 21094 (Thesis). ETH Zurich, Zurich, Switzerland.
https://www.research-collection.ethz.ch/handle/20.500.11850/69139
* Quaglia, I. et al. (2024): The potential of stratospheric aerosol injection to reduce the climatic risks of explosive volcanic eruptions. Geophysical Research Letters. https://doi.org/10.1029/2023GL107702
* Richter, J. H. et al. (2022): Assessing responses and impacts of solar climate intervention on the Earth system with stratospheric aerosol injection (ARISE-SAI). Geoscientific Model Development. https://doi.org/10.5194/gmd-15-8221-2022
* Samset, B. H. et al. (2025): East Asian aerosol cleanup has likely contributed to the recent acceleration in global warming. Communications Earth & Environment. https://doi.org/10.1038/s43247-025-02527-3
* Tang, A. et al. (2021): A fate worse than warming? Stratospheric aerosol injection and global catastrophic risk. Frontiers in Climate. https://doi.org/10.3389/fclim.2021.720312
* Trisos, C. H. et al. (2018): Potentially dangerous consequences for biodiversity of solar geoengineering implementation and termination. Nature Ecology & Evolution. https://doi.org/10.1038/s41559-017-0431-0
* Visioni, D. et al. (2023a): Opinion: The scientific and community-building roles of the Geoengineering Model Intercomparison Project (GeoMIP). Atmospheric Chemistry and Physics. https://doi.org/10.5194/acp-23-5149-2023
* Visioni, D. et al. (2023b): Climate response to off-equatorial stratospheric sulfur injections in three Earth system models. Atmospheric Chemistry and Physics. https://doi.org/10.5194/acp-23-663-2023
* Visioni, D. et al. (2024): G6-1.5K-SAI: A new Geoengineering Model Intercomparison Project (GeoMIP) experiment integrating recent advances in solar radiation modification studies. Geoscientific Model Development. https://doi.org/10.5194/gmd-17-2583-2024
* Wiertz, T. (2015): Visions of climate control: Solar radiation management in climate simulations. Science, Technology, & Human Values. https://doi.org/10.1177/0162243915606524
-
AC3: 'Response to Ben Kravitz', Benjamin Sanderson, 04 Mar 2026
-
RC2: 'Comment on egusphere-2026-28', Gideon Futerman, 26 Feb 2026
This paper is, in my mind, trying to present a perspective on a rather significant question in the assessment of SRM: how we can integrate the non-ideal dynamics of SRM governance into SRM scenario generation. Given both how significant this question is, and the increasing discussion in the literature on it, a discussion presenting the building blocks of an alternative approach, or building on the present state of the literature, would be very timely.
At its core, I like what the paper is trying to do. Creating novel modelling frameworks that let us make progress on better exploring different futures that consider the full range of socio-political dynamics is an important problem in the field. Indeed, I am less optimistic than other reviewers that the SRM modelling community is making the right trade-offs in this regard at present.
However, I cannot accept the paper as it is. I think the authors can either be taken as making strong claims, which, whilst I believe could be supported, are not here adequately argued for in the paper. Or they can be taken to be pointing out particular flaws in current modelling practices that have generally been pointed out by other authors before. Their proposed architecture, to move beyond previous literature, must be more practical than the current sketch.
This paper does not do an adequate job at engaging with the existing literature on SRM and non-ideal scenarios. I agree with the authors that existing modelling practices mostly use idealised scenarios and have mostly failed to adequately model concerning non-ideal scenarios. However, the failure to engage with the literature leaves the reader with an importantly distorted impression. Firstly, there have been a number of attempts made to engage with more realistic non-ideal scenarios in recent years. Secondly, modellers have generally made the choice to model SRM in a specific way for specific reasons, some stated and some implicit.
The second issue I have is with the supposed subject of the paper. It is never really explained as to what “robust” assessment of SRM entails. What is it that makes an assessment robust, and how high a bar are we drawing? These need to be laid out by the authors. I think there is a strong argument to make that the authors are correct in their claim; however, engagement with what robust risk assessments need to entail is needed to make such a claim. I would also recommend looking into the debates around Global Catastrophic Risk in the context of both SRM and climate change, who have been making similar arguments to this paper for a while.
This argument in fact needs to go further. The paper claims it shows “what must be represented for adequate SRM assessment,”; the authors, if they wish to make this claim, need to show not only that “robust” assessments of SRM must include these shocks but “adequate” (for what purpose?) assessments must as well. This requires laying out the criteria for adequacy of assessment, and being explicit about the value judgements used in such an assessment.
I do think such an argument could be made. I think there are many ways to make it, and I will briefly sketch one out that I am sympathetic to, but I am sure there are similar arguments that the authors could make. Kemp et al. 2022 makes the argument that consideration of Global Catastrophic Risk in the context of climate change is needed for truly adequate risk assessment. Then, works on SRM and GCR (eg Halstead 2018, Futerman and Beard, 2023) have made the argument that the “fast loop dynamics” are the key drivers of Global Catastrophic Risk from SRM. If these claims are taken together, it would suggest that without fast loop assessments, then truly adequate risk assessments cannot be done.
Further to this, for this paper to be actionable, it probably needs to make the argument that the trade offs made by the modelling community that have led them to the current modelling approach are not the correct trade offs to make. Namely, they have to argue that moving closer to what notion of “robust” or “adequate” risk assessment of SRM the authors choose is the direction that the field should move in, even at the cost of less resources devoted to other modelling activities. (Whilst not strictly needed for the paper to make sense, given how normatively loaded this is, such an argument seems important). This paper fails to do this, instead, recapitulating the various (valid) issues of current modelling frameworks. A paper that dove into weighing up when and why we should prioritize the sorts of questions that this approach aids us in would be a far more helpful contribution than the paper currently is, and would move the paper much closer to being publishable.
Another approach the authors could take is to scale down the ambition of their claims, to merely providing a framework that could guide the modelling of non-ideal scenarios. Whilst it is incumbent on them in this case to show that this usefully differs from current proposed approaches, I think they could do that. However, if they were to reduce the ambition of their claims, it appears to me they would need to add significantly to this paper in order for it to be novel enough to publish. Whilst previous papers on non-ideal scenarios don’t contain every element in this paper, enough is similar that I don’t think this paper would cross the novelty bar for publication. With more detailed discussions of how to practically actualise the “fast loop” than the current brief sketch provides, this paper would provide a useful contribution to the literature. Essentially, an expansion of Section 5 that begins to show how the framework is actionable would be what needs to be added to the paper. My view is that the core of the paper should essentially be showing how different approaches in section 5 could be applied to the developed architecture.
On the other side, I do rather like the formalizations present in the paper, and do think the architecture could be the foundation for some very important progress. However, I think this needs to be more fleshed out and actioned before it is publishable. Similarly, I think the argument that such an architecture is needed is important (especially from a GCR perspective), but think the argument needs to be stronger before this is publishable.
Specific Comments:
135-150 I’m basically not sure the argument really works. I did find it quite hard to parse, but it basically seemed to compare SRM to ideal CDR scenarios with purely rational, climate focused, decisionmakers, whereas much of the point of the fast loop is to eliminate some of these idealising features. More broadly, it is not clear to me why the Moral Hazard is in the “Fast Loop” under SRM, but not for CDR scenarios. This seems fairly at odds with the literature on moral hazard, which makes far less of a distinction between moral hazard generated from CDR and SRM.
178: “purely unilateral” It is strange to assume unilateral deployment must be regional. Many discussions of unilateral deployment may assume limited injection sites, but this doesn’t necessarily mean regional deployment
186: I think a number of important to consider scenarios that seem to be unable to be modelled under the operational rules. Counter-geoengineering is a key one. Whilst an argument could be made for including this under Operational Inefficiency, the impacts of counter-geoengineering may be sufficiently difficult that I am unsure that rule will be able to adequately capture the scenarios. Similarly, deployment type is not included, despite the fact that whether it is MCB or SAI or a cocktail of both is both absolutely essential for modelling and may be impacted by the Fast Loop processes. Finally, whilst “Deployment Regionality” covers many scenarios, scenarios where there are multiple deployments, each individually optimized but no joint optimization, is also not discussed. I would recommend revisiting the Operational Rules to allow them to better characterise different types of SRM deployment.
187: Similarly, important scenarios are not considered in Context Loop Properties. Especially termination relevant factors seem to be strongly affected by Societal Vulnerability to Catastrophic Shocks (Baum et al 2012) with a Global Catastrophic Shock amongst the more likely causes of termination. Such shocks do not happen in isolation, they are dependent on the context in which they impact, which is in turn dependent on the rest of the scenario. However, such vulnerability is not included at present in the Context Loop Properties.
364: I think you should at least discuss decision-theoretic frameworks. These are a very important part of robust assessments, and may become even more significant if this approach makes systematic scenario generation harder. This paper seems to really suggest a move in a particular direction for modelling scenarios, towards sets of scenarios where normal multi-model ensembles and systematic assessment is harder. If this is so, then guidance must be included as to what comes next.
379: “ The architecture we have outlined here is not a solution to this problem; it is an attempt to highlight the scope of the risk assessment challenge.”
My assessment is that this is not a unique contribution to the literature in its current form because of this. Other papers have laid out the difficulty of robust risk assessment of SRM from current models, using similar arguments to yours. For this paper to be novel, it must actually start to move towards a solution to the problem.Citation: https://doi.org/10.5194/egusphere-2026-28-RC2 -
AC2: 'Response to Gideon Futerman', Benjamin Sanderson, 04 Mar 2026
Many thanks to the reviewer for a very constructive review. The Global Catastrophic Risk framing is highly relevant and we’ve made efforts to adopt it centrally in revision. We’ve also worked to clarify the practical implications of our framework - outlining more real-world use cases for modeling the different SMRPs. We respond point-by-point below.
General concern 1: Literature engagement — GCR and SRM
"This paper does not do an adequate job at engaging with the existing literature on SRM and non-ideal scenarios. [...] there have been a number of attempts made to engage with more realistic non-ideal scenarios in recent years."
We agree with this critique - which was raised also by other reviewers, and the revised manuscript now engages far more deeply with the relevant literature. We have added 43 additional citations spanning physical SRM modelling, governance and social science, climate-society feedbacks, and — prompted specifically by this reviewer — the Global Catastrophic Risk literature. We have reworked the introduction to follow a four-part logical progression: (1) MIPs explore physical responses; (2) policy-relevant futures require IAM/SSP frameworks whose enabling assumption is slow GHG timescales; (3) the SSP framework does not incorporate shocks, by design; (4) SRM breaks this timescale assumption. A new paragraph on social science and governance work, and a gap analysis mapping existing studies against the SRMP typologies, are now included.
The new literature falls into five clusters, weighted toward the topics emphasised by the reviewer. The GCR literature now anchors the revised framing, led by Kemp et al. (2022), Halstead (2018), and Futerman and Beard (2023) on fast-loop governance dynamics as the key catastrophic risk mechanism. We now engage substantively with the DMDU (Decision-making under deep uncertainty) literature, including the only existing application of robust decision-making to geoengineering governance (Lempert and Prosnitz, 2011). The moral hazard treatment now draws on mixed empirical evidence and conceptual work spanning six studies. Social climate models (Beckage et al., 2022, 2025) and emulator-based approaches (Farley et al., 2024, 2026) provide the technical backbone for the restructured Figure 5. The remaining citations extend coverage of SRM physical modelling and scenario design.
The full list of the 43 additional citations, organised by all topics across all three reviews, appears in our response to Dr. Visioni.
________________
General concern 2: What does "robust" or "adequate" assessment mean?
"It is never really explained as to what 'robust' assessment of SRM entails. What is it that makes an assessment robust, and how high a bar are we drawing?"
"The authors, if they wish to make this claim, need to show not only that 'robust' assessments of SRM must include these shocks but 'adequate' (for what purpose?) assessments must as well. This requires laying out the criteria for adequacy of assessment..."
Thanks for this point. We now define three complementary dimensions of robustness:
"Our use of 'robust assessment' refers to a complementary concern on two levels. The first is process completeness: whether the assessment framework represents the classes of dynamics most relevant to the risk problem. [...] The second is coverage adequacy: whether the space of futures explored is sufficient to characterise the risk landscape, given the processes represented. An assessment can be model-robust (well-validated across independent model constructions) yet process-incomplete (lacking the coupled feedback mechanisms that generate primary policy-relevant risks) and coverage-limited (unable to explore the futures those processes would produce). All three dimensions of robustness (model validation, process representation, and coverage) are necessary; none alone is sufficient. The SRMP framework addresses primarily the second and third dimensions, while acknowledging and building on the first."
On the "adequate for what?" question, the revised manuscript now provides an explicit answer:
"If the dominant channels through which SRM contributes to catastrophic outcomes are fast-loop dynamics (governance failure, geopolitical disruption, termination under societal stress; (Tang 2021, Futerman & Beard 2023), then an assessment that cannot represent these dynamics is, by construction, inadequate for the decisions it is intended to inform, regardless of its model-robustness on other dimensions."
________________
General concern 3: GCR framing
"I would also recommend looking into the debates around Global Catastrophic Risk in the context of both SRM and climate change, who have been making similar arguments to this paper for a while."
"Kemp et al. 2022 makes the argument that consideration of Global Catastrophic Risk in the context of climate change is needed for truly adequate risk assessment. Then, works on SRM and GCR (eg Halstead 2018, Futerman and Beard, 2023) have made the argument that the 'fast loop dynamics' are the key drivers of Global Catastrophic Risk from SRM."
We have adopted this framing directly. A new paragraph in the Introduction now engages the GCR literature:
"A growing literature examines SRM through the lens of global catastrophic risk (GCR), identifying termination shock, interaction with other catastrophic hazards, and governance fragility as primary channels through which SRM could contribute to catastrophic outcomes (Halstead, 2018; Tang and Kemp, 2021; Kemp et al., 2022; Futerman and Beard, 2023). These analyses converge on a central finding: SRM's most severe risks arise not from its steady-state physical effects but from fast-loop dynamics (loss of deployment capacity, geopolitical disruption, cascading system failures) that existing scenario frameworks are structurally unable to resolve. The framework proposed here responds directly to this challenge."
This has become a highly relevant thread in the revised paper. Additionally, the reviewer's recent work on SRM and Earth system tipping elements (Futerman et al., 2025) is now cited in the introduction, in the context of non-ideal SRM dynamics that have been explored in the recent literature.
________________
General concern 4: Opportunity cost and trade-offs
"they have to argue that moving closer to what notion of 'robust' or 'adequate' risk assessment of SRM the authors choose is the direction that the field should move in, even at the cost of less resources devoted to other modelling activities."
Good point. A new paragraph in Section 5 addresses this:
"We acknowledge that broadening assessment beyond idealised experiments involves resource and attention trade-offs. However, these trade-offs are less zero-sum than they may appear: the disciplinary communities best placed to represent governance dynamics, political shocks, and socioeconomic feedbacks are largely distinct from those running Earth system experiments, and many of these fields, including integrated assessment, political science, and disaster-risk research, must grapple with coupled human-Earth system feedbacks regardless if their own frameworks are to remain relevant in an era where large-scale intervention is actively debated. The SRMP framework sharpens rather than multiplies this burden: by making the gap structure explicit, it identifies where additional representational capacity would yield the greatest marginal reduction in decision-relevant uncertainty, enabling targeted investment rather than diffuse expansion."
________________
General concern 5: Novelty and actionability of Section 5
"For this paper to be novel, it must actually start to move towards a solution to the problem."
"an expansion of Section 5 that begins to show how the framework is actionable would be what needs to be added to the paper. My view is that the core of the paper should essentially be showing how different approaches in section 5 could be applied to the developed architecture."
This was the most substantial revision. Section 5 has been restructured around a new Figure 5, which replaces the original monolithic pipeline diagram with four panels showing typology-specific assessment architectures:
Updated Figure 5 — four-panel typology-specific suggested modeling framework
This section now accompanies the revised figure:
“A central implication of the SRMP typologies is that different governance logics require structurally different modelling ecosystems. Figure 5 illustrates this by mapping each typology family to a minimal assessment architecture across three shared dimensions (Governance, Earth System, and Socio-economic context), showing which loops are active and which tool classes fill each role. The consistent layout makes visible both the shared Earth System component across all typologies and the structural absence of socio-economic feedback in current approaches (Panel A).
Panel A: SRMP-a/b (Idealised and Imperfect Cooperation). This architecture corresponds to the existing assessment pipeline. A feedback controller \citep{Kravitz2017} modulates injection rates in response to climate output from an ESM or emulator, with governance treated as a fixed exogenous boundary condition and emissions provided by a conventional IAM. The Dynamic Context Loop is inactive: governance is assumed stable and cooperative, so there is no mechanism for shocks to alter the socio-political context. GeoMIP experiments \citep{Kravitz2015}, ARISE-SAI \citep{Richter2022, Henry2023}, and emulator-based scenario exploration \citep{Farley2024, Farley2026} all operate within this architecture. It provides essential physical climate understanding but cannot represent the governance-mediated risks that define SRMP-c through SRMP-e.
Panel B: SRMP-c (Conflict). The key architectural change is in the Fast Loop: the single cooperative controller is replaced by an N-agent strategic model (game-theoretic \citep{Pezzoli2023} or agent-based \citep{Castro2020}) in which multiple actors choose injection strategies according to competing objectives. The emulator translates superposed strategies into regional climate responses, filtered through detection/attribution uncertainty: actors respond to \textit{perceived} impacts, and misattribution can trigger escalation independent of actual climate change. The Dynamic Context Loop is now active: conflict events degrade a cooperation index and increase societal vulnerability \citep{Baum2013}, which in turn modulates Interruption Risk and the severity of subsequent shocks. This creates the positive feedback between shock occurrence and shock susceptibility discussed in Section~4.
Panel~C: SRMP-d (Free Driver / Inequality). Here the dominant feedback pathway runs through regional equity. A coalition-optimised deployment produces spatially heterogeneous climate impacts; the emulator resolves these into distributional outcomes that are evaluated through equity perception metrics. Perceived inequity feeds back into governance fragmentation (as affected populations or their representatives respond to unequal outcomes), which modifies the coalition's capacity and willingness to sustain deployment. The modelling requirement is distinct from Panel~B: rather than strategic interaction among deployers, the critical dynamic is the distributional consequences of a single coalition's choices feeding back through governance instability.
Panel~D: SRMP-e (Moral Hazard). The architecture here centres on a self-reinforcing substitution cycle. Technically successful SRM deployment produces a perceived-success signal that erodes mitigation ambition. Reduced ambition shifts the emissions trajectory, which the emulator translates into a modified joint climate outcome (SRM forcing plus higher baseline warming). The resulting increased dependence on SRM reinforces continued deployment and further weakens mitigation incentives. The key modelling requirement is an extended IAM (or equivalent) capable of representing endogenous mitigation-ambition response to SRM outcomes, rather than treating emissions as an exogenous pathway. Stochastic policy events (elections, economic shocks) modulate the ambition response, introducing variability into the substitution dynamic.”
Critically, the emulator is positioned as a shared enabling technology across all four panels, and the text notes that recent work on emulating SAI inconsistencies (Farley et al., 2024) and developing climate-intervention dynamical emulators for scenario-space exploration (Farley et al., 2026) demonstrates the feasibility of this approach.
The new Section 5 also opens with the key insight that different typologies require structurally different modelling ecosystems:
"A central implication of the SRMP typologies is that different governance logics require structurally different modelling ecosystems. Figure 5 illustrates this by mapping each typology family to a minimal assessment architecture, showing which loops are active and which tool classes fill each role."
________________
General concern 6: Decision-theoretic frameworks (line 364)
"I think you should at least discuss decision-theoretic frameworks. These are a very important part of robust assessments, and may become even more significant if this approach makes systematic scenario generation harder."
Agreed. The revised manuscript now engages substantively with DMDU:
"The field of Decision Making under Deep Uncertainty (DMDU) offers a family of alternatives (Marchau et al., 2019): robust decision-making seeks strategies that perform acceptably across a wide range of plausible futures rather than optimally for any single forecast (Weaver et al., 2013); scenario discovery uses computational search to identify the combinations of uncertain parameters that most sharply distinguish policy success from failure (Bryant and Lempert, 2010); and dynamic adaptive policy pathways sequence near-term actions with pre-defined triggers for course correction. These approaches have already gained traction in IPCC assessment, particularly for adaptation and sea-level-rise planning (Lempert et al., 2024), and Lempert and Prosnitz (2011) demonstrated a preliminary application of robust decision-making to geoengineering governance."
The key argument is that DMDU methods require a structured characterisation of the uncertainty space as input, and the SRMP typologies provide exactly this:
"Critically, all DMDU methods require as input a structured characterisation of the relevant uncertainty space: the dimensions along which futures can diverge and the mechanisms that drive divergence. The SRMP typologies provide exactly this, mapping each governance logic to a distinct set of uncertain parameters, feedback pathways, and failure modes. A natural next step is therefore to apply scenario-discovery techniques to the typology-specific architectures of Figure 5, identifying the parameter combinations under which each SRMP pathway crosses thresholds of unacceptable harm."
The "Economic frameworks for stochastic catastrophic risk" bullet in the Priority Research Domains now also cites the DMDU literature specifically, including Lempert and Prosnitz (2011) as the only existing application of robust decision-making to geoengineering governance. The conclusion has been updated to reframe the evaluation gap from "undefined" to an achievable research frontier: DMDU evaluation methods exist, the SRMP typologies provide the structured input they require, and connecting the two is a tractable next step.
________________
Specific comment: Lines 135–150 (Moral hazard and CDR)
"I'm basically not sure the argument really works. [...] it is not clear to me why the Moral Hazard is in the 'Fast Loop' under SRM, but not for CDR scenarios."
Fair challenge. The distinction is not that moral hazard is absent for CDR but that its timescale and coupling structure differs. For CDR, a reduction in mitigation ambition feeds back through slow GHG accumulation; the climate signal from reduced CDR takes decades to manifest, and the governance response to that signal is equally slow. For SRM, the feedback is fast: deployment success is perceived on sub-decadal timescales, mitigation ambition can respond within political cycles, and the climate consequence of reduced mitigation (continued dependence on SRM) is locked in almost immediately. This is a tight loop, not a slow drift.
Beyond this, the Mitigation Delay property is designed to be bidirectional: it can take positive values (mitigation deterrence) or negative values (spurring ambition). The empirical literature is genuinely mixed: Andrews et al. (2022), Cherry et al. (2023), Abatayo et al. (2020), and Acemoglu and Rafey (2018) are all now cited. The framework does not presuppose which direction this coupling takes.
________________
Specific comment: Line 178 (Unilateral does not mean regional)
"It is strange to assume unilateral deployment must be regional. Many discussions of unilateral deployment may assume limited injection sites, but this doesn't necessarily mean regional deployment."
We agree with this point from the reviewer. We've rewritten the Deployment Regionality definition:
"Deployment Regionality: The scope of the optimisation objective governing the deployment design. This property defines whose climate outcomes the intervention is designed to serve: globally equitable objectives, a specific coalition's interests, or a single actor's national priorities. Deployment Regionality is a governance property, not an injection geography: a unilateral deployer may inject globally while optimising for national interests, and a cooperative deployment may target a specific region by mutual agreement."
SRMP-c now reads "high Deployment Regionality (self-interest optimised, whether injection is global or regional)" and SRMP-d reads "coalition-interest optimised." Figure 3 has been updated to reflect governance scope rather than injection geography.
________________
Specific comment: Line 186 (Operational Rules gaps)
"I think a number of important to consider scenarios that seem to be unable to be modelled under the operational rules. Counter-geoengineering is a key one. [...] Similarly, deployment type is not included [...] Finally, [...] scenarios where there are multiple deployments, each individually optimized but no joint optimization, is also not discussed."
Thanks for these point - we address them with three changes:
Counter-geoengineering and multiple deployers. A new paragraph on "Emergent compound dynamics" describes counter-geoengineering as a high-severity realisation within the existing property space where Interruption Risk, D/A Uncertainty, and Conflict Potential simultaneously reach extreme values, producing qualitatively different interaction effects (adversarial interference, attribution collapse, escalation spirals). Multiple uncoordinated deployers are treated as a generalisation of SRMP-c logic: several self-interest-optimised actors producing superposed forcings with incoherent net effects.
"Counter-geoengineering, defined as deliberate intervention by one actor to offset or neutralise another's deployment, emerges naturally as a high-severity realisation within SRMP-c: it simultaneously elevates Interruption Risk (deployment effectiveness becomes contested), degrades Detection/Attribution (overlapping interventions with opposing objectives create signals that may be practically unattributable), and intensifies Conflict Potential beyond bilateral disagreement into active adversarial interference. Similarly, multiple uncoordinated deployers with conflicting objectives represent a generalisation of the SRMP-c logic in which several actors each follow self-interest-optimised strategies concurrently, producing superposed climate forcings whose net effect may be incoherent and whose attribution to individual actors becomes intractable."
Deployment type (SAI vs MCB). A new subsection, "Deployment mechanism as a conditional modulator," with an accompanying Table 1, addresses this substantively. The argument is that deployment mechanism is not an independent governance dimension (not an 8th Driver Property) but a conditional modulator that shifts the feasible range, probability, and physical consequences of all seven existing properties simultaneously. Table 1 maps SAI and MCB across all seven Driver Properties (e.g., termination shock vs. intermittent disruption for Interruption Risk; global vs. naturally regional for Deployment Regionality; well-studied vs. uncertain at scale for Operational Inefficacy).
"Rather, deployment mechanism acts as a conditional modulator that shifts the feasible range, probability, and physical consequences of all seven existing properties simultaneously. [...] The framework therefore provides a mechanism-general governance-risk language. The five typologies and seven properties describe governance logics that apply to any fast-responding SRM technique; mechanism-specific instantiation then determines which regions of the property space are physically realisable, how probable each typology is, and what the resulting climate trajectories look like."
________________
Specific comment: Line 187 (Vulnerability to catastrophic shocks)
"Especially termination relevant factors seem to be strongly affected by Societal Vulnerability to Catastrophic Shocks (Baum et al 2012) with a Global Catastrophic Shock amongst the more likely causes of termination. Such shocks do not happen in isolation, they are dependent on the context in which they impact..."
We absolutely agree with this, and it’s what we were trying to capture with the context feedbacks - but in revision we’ve tried to make the potential for these vulnerability shifts clearer. A new paragraph in the State section addresses this:
"Critically, societal vulnerability to catastrophic shocks is not a fixed background condition but a dynamic property of the State that co-evolves with the SRM trajectory itself (Figure 2). As Baum et al. (2013) argue, SRM termination is most likely to occur precisely when it would be most damaging: during a global catastrophic shock (pandemic, financial crisis, major conflict) that simultaneously degrades the institutional capacity needed to maintain deployment and increases the population's exposure to termination-driven climate rebound. Each shock that degrades governance capacity, institutional trust, or economic resilience raises vulnerability to subsequent shocks, creating a positive feedback within the Dynamic Context Loop."
Figure 2 has been updated to show modified vulnerability after context-changing events.
________________
Specific comment: Line 379 ("Not a solution")
"My assessment is that this is not a unique contribution to the literature in its current form because of this. Other papers have laid out the difficulty of robust risk assessment of SRM from current models, using similar arguments to yours. For this paper to be novel, it must actually start to move towards a solution to the problem."
Thanks for the clear assessment. We’ve worked to clarify the actionability of our framework and believe the revised manuscript now provide much more pragmatic guidance for future modeling and assessment. The combination of (a) the GCR-grounded adequacy argument, (b) the formal typology with seven bounded Driver Properties and five governance-logic typologies, (c) the four-panel Figure 5 showing typology-specific assessment architectures with identified tool classes and coupling mechanisms, (d) the substantive treatment of deployment mechanism as a conditional modulator (Table 1), and (e) the connection to DMDU evaluation frameworks constitutes a significantly more actionable contribution than the original submission. The paper now identifies not only what must be represented but how each typology maps to structurally distinct modelling ecosystems, and how the resulting scenario ensembles can be evaluated under deep uncertainty.
The revised conclusion reflects this shift:
"The DMDU community has developed evaluation methods for precisely such conditions (Marchau et al., 2019; Lempert et al., 2024), and the SRMP typologies provide the structured uncertainty space these methods require as input. Connecting the two is an achievable and essential research frontier."
________________
We hope the revision makes clear how much this review shaped the paper. The GCR framing, the adequacy criteria, the DMDU connection, and the restructured Figure 5 all trace directly to points raised here.
________________
References cited in this response
* Abatayo, A. L. et al. (2020): Solar geoengineering may lead to excessive cooling and high strategic uncertainty. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.1916637117
* Acemoglu, D. and Rafey, W. (2018): Mirage on the horizon: Geoengineering and carbon taxation without commitment. NBER Working Paper No. 24411. https://doi.org/10.3386/w24411
* Andrews, T. M. et al. (2022): Anticipating moral hazard undermines climate mitigation in an experimental geoengineering game. Ecological Economics. https://doi.org/10.1016/j.ecolecon.2022.107421
* Baum, S. D. et al. (2013): Long-term trajectories of human civilization. Foresight. https://doi.org/10.1108/14636681311321121
* Beckage, B. et al. (2022): Incorporating human behaviour into Earth system modelling. Nature Human Behaviour. https://doi.org/10.1038/s41562-022-01478-5
* Beckage, B. et al. (2025): Models and scenarios for solar radiation modification need to include human perceptions of risk. Environmental Research: Climate. https://doi.org/10.1088/2752-5295/addd42
* Brody, E. et al. (2024): Kicking the can down the road: understanding the effects of delaying the deployment of stratospheric aerosol injection. Environmental Research: Climate. https://doi.org/10.1088/2752-5295/ad53f3
* Bryant, B. P. and Lempert, R. J. (2010): Thinking inside the box: A participatory, computer-assisted approach to scenario discovery. Technological Forecasting and Social Change. https://doi.org/10.1016/j.techfore.2009.08.002
* Cherry, T. L. et al. (2023): Climate cooperation in the shadow of solar geoengineering: An experimental investigation of the moral hazard conjecture. Environmental Politics. https://doi.org/10.1080/09644016.2022.2066285
* Diao, C. et al. (2025): A model study exploring the decision loop between unilateral stratospheric aerosol injection scenario design and Earth system response. Earth's Future. https://doi.org/10.1029/2024EF005455
* Farley, A. et al. (2024): Emulating inconsistencies in stratospheric aerosol injection. Environmental Research: Climate. https://doi.org/10.1088/2752-5295/ad519c
* Farley, A. et al. (2026): A Climate Intervention Dynamical Emulator (CIDER). Geoscientific Model Development. https://doi.org/10.5194/egusphere-2025-1830
* Futerman, G. et al. (2025): The interaction of solar radiation modification with Earth system tipping elements. Earth System Dynamics, 16, 939–978. https://doi.org/10.5194/esd-16-939-2025
* Futerman, L. and Beard, S. (2023): Geoengineering and global catastrophic risk. Futures. https://doi.org/10.1016/j.futures.2023.103206
* Halstead, J. (2018): Stratospheric aerosol injection research and existential risk. Futures. https://doi.org/10.1016/j.futures.2018.03.004
* Jebari, J. et al. (2021): From moral hazard to risk-response feedback. Climate Risk Management. https://doi.org/10.1016/j.crm.2021.100324
* Jones, A. et al. (2013): The impact of abrupt suspension of solar radiation management (termination effect) in experiment G2 of the Geoengineering Model Intercomparison Project (GeoMIP). Journal of Geophysical Research: Atmospheres. https://doi.org/10.1002/jgrd.50762
* Kemp, L. et al. (2022): Climate Endgame: Exploring catastrophic climate change scenarios. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.2108146119
* Kravitz, B. et al. (2015): The Geoengineering Model Intercomparison Project Phase 6 (GeoMIP6): Simulation design and preliminary results. Geoscientific Model Development. https://doi.org/10.5194/gmd-8-3379-2015
* Lempert, R. J. et al. (2024): Decision making under deep uncertainty for climate adaptation. Annual Review of Environment and Resources. https://doi.org/10.1146/annurev-environ-112321-095025
* Lempert, R. J. and Prosnitz, D. (2011): Governing geoengineering research: A political and technical vulnerability analysis of potential near-term options. RAND Corporation. https://doi.org/10.7249/TR846
* Lloyd, E. A. (2010): Confirmation and robustness of climate models. Philosophy of Science. https://doi.org/10.1086/657427
* Lockley, A. et al. (2022): Politically relevant solar geoengineering scenarios. Socio-Environmental Systems Modelling. https://doi.org/10.18174/sesmo.18127
* MacMartin, D. G. et al. (2022): Scenarios for modeling solar radiation modification. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.2202230119
* Marchau, V. A. W. J. et al. (2019): Decision Making under Deep Uncertainty: From Theory to Practice. Springer. https://doi.org/10.1007/978-3-030-05252-2
* McLaren, D. (2016): Mitigation deterrence and the "moral hazard" of solar radiation management. Earth's Future. https://doi.org/10.1002/2016EF000445
* Pflüger, D. et al. (2024): Flawed emergency intervention: Slow ocean response to abrupt stratospheric aerosol injection. Geophysical Research Letters. https://doi.org/10.1029/2023GL106132
* Tang, A. et al. (2021): A fate worse than warming? Stratospheric aerosol injection and global catastrophic risk. Frontiers in Climate. https://doi.org/10.3389/fclim.2021.720312
* Trisos, C. H. et al. (2018): Potentially dangerous consequences for biodiversity of solar geoengineering implementation and termination. Nature Ecology & Evolution. https://doi.org/10.1038/s41559-017-0431-0
* Visioni, D. et al. (2024): G6-1.5K-SAI: A new Geoengineering Model Intercomparison Project (GeoMIP) experiment integrating recent advances in solar radiation modification studies. Geoscientific Model Development. https://doi.org/10.5194/gmd-17-2583-2024
* Weaver, C. P. et al. (2013): Improving the contribution of climate model information to decision making: the value and demands of robust decision frameworks. Wiley Interdisciplinary Reviews: Climate Change. https://doi.org/10.1002/wcc.202
-
AC2: 'Response to Gideon Futerman', Benjamin Sanderson, 04 Mar 2026
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 963 | 360 | 21 | 1,344 | 25 | 23 |
- HTML: 963
- PDF: 360
- XML: 21
- Total: 1,344
- BibTeX: 25
- EndNote: 23
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The first comment that I have for this piece is that the authors, for some reason, have decided to complete ignore much of the conversation already happening in SRM research about scenarios, and essentially decided to build a strawman of it to criticize, instead of fairly reviewing already existing discussions, engaging with them and only at that point, of course, criticizing them, which is fine as long as this is done on the basis of what past research actually says. I will note that I already made this point when I reviewed another version of this piece for another journal, where it was ultimately not accepted, but there have been no advances by the authors on this aspect. I can only assume they do not find the idea of engaging with recent literature valuable, so, if they persist, at least being this a public forum, I am hoping a reader would also find my comment valuable for added context. As a side note, my review will be signed, and I don’t want to make a mystery of the fact that (as the co-chair of GeoMIP) I was involved in much of this scenario work the authors decided to ignore.
Let’s start with this phrase in the introduction:
“This leaves the most critical, policy-relevant risks of SRM, such as governance failure and geopolitical conflict (Pezzoli et al., 2023; Cherry et al., 2024), abrupt termination (Parker and Irvine, 2018), and regional impact disparities (Heyen et al., 2015) un-modeled by either ESM or IAM paradigms.”
This is just not true. MacMartin et al. (2022) simulated a broad range of scenarios, including termination, phase-out and multiple levels of cooling, and discussed the importance (and challenges) of assessing a broad range of them, as well, and the termination was also a part of one of the initial GeoMIP experiment, G4, which were analyzed for instance in Trisos et al. (2018), and will be a part of the CMIP7 GeoMIP scenarios. Quaglia et al. (2024) further expanded the space to volcanic eruptions happening during SAI (already explored in Laakso et al., 2016), and Brody et al. (2024) and Pfluger et al. (2024) analysed scenarios with delayed deployments and, especially the latter, focused on how it would fail to revert some key tipping dynamics. Very recently, Estrada et al. (2026) also discusses this at length.
More broadly, unlike what the authors say in the conclusions:
“This is why naïve assessment built on idealised assumptions poses such acute risks at this juncture. If the research community evaluates SRM primarily through “best-case” scenarios – smooth deployment, perfect cooperation, orderly termination – it provides a systematically distorted evidence base for decisions that could prove irreversible.”
The discussion of how much and when to simulate non-ideal scenarios was often discussed in GeoMIP, which the authors do not acknowledge. For instance, see the review in Visioni et al. (2023a) (see Section 6.1.1, dedicated explicitly to this, and emerging from very long discussions within the community), and in Visioni et al. (2024) (which the authors cite, but did not bother reading, because the reasoning behind the scenario choice were discussed with some attention, rather than being naive, which in the english language means “a lack of experience, wisdom, or judgment”). I’ll note that both works were the results of years of engagement with the community, talks by experts about scenario generation, evidence gathered from past studies and suggestions from modeling groups, impact analyses groups and governance experts (see the yearly meeting reports here: https://climate.envsci.rutgers.edu/GeoMIP/publications.html#np). Now, it was because, as we laid out in many of these work, it is hard to have coordinate, multi-model experiments that are both able to shine a light on the climatic uncertainties and can meaningfully span the scenario space, in Visioni et al. (2023a) there was an ample discussion of the potential to use more idealized simulations to train climate emulators that would be able to better span the space, instead. Hence the line of work in Farley et al. (2024) and Farley et al. (2026) to develop such an emulator that is capable both of exploring the space of interruptions and terminations (Farley et al., 2024) and that of uncoordinated deployments and regional impacts (Farley et al., 2026) using a broader range of idealized simulations described in the GeoMIP testbed experiments in Visioni et al. (2023b). There is other literature on such non-optimal scenarios including single-case models simulations, as well, which demonstrate the community is indeed not leaving this aside: see Diao et al. (2025) for a case of unilateral action, Wan et al. (2024) for a case using MCB, Kwiatkowski et al. (2015) for another “edge” case of intervention, Jackson et al. (2015) for a “real world” implementation of SAI targeted at restoring sea-ice considering imperfect observations, etc. But what made those cases useful to analyse and possible to model was the work of more idealized or regular simulations that allowed for models’ and scenarios improvement, and diagnosis of uncertainty.
Lastly, the authors also ignored large pieces of the literature in realms different from just climate science, curiously forgetting also the ones that offer various fair criticism of current scenarios, but that at least try to engage in the broader discourse rather than sidestep it. See for instance Wiertz et al. (2015), McLaren and Corry (2021), Buck (2022), Keys (2023), Beckage et al. (2025) (which explicitly talks about a lot of the “societal feedbacks” in this work’s title…) and much more.
Now, I could probably continue on this line, but I want to focus on why I made (again) the list of works this list of authors missed. A perspective piece is not a review piece, and yet it must fairly address others’ perspectives before engaging in new proposals. The authors fail to do so, and given my previous feedback, I can’t assume due to naivete on their part, but due to a deliberate choice to erase other perspectives. If the authors wanted to engage, they would describe how their proposed framework integrates them, rather than ignore them.
Moving on to my second point, the authors often repeat throughout the piece that any other framework but theirs is wholly inadequate, naive, etc. mainly because there is something, in SRM, that makes this kind of exploration of the broader set of scenarios they propose conditio sine qua non for for any robust assessment. It would be interesting to know how the authors define “robustness”. Personally, and also in the context of IPCC assessments, robustness is something I identify with perspectives such as Lloyd (2010, 2015) related to climate models’ robustness: in her words robustness is the “repeated production of the empirically successful model prediction or retrodiction against a background of independently-supported and varying model constructions”. In this sense, a robust SRM assessment is one where the tools used for the assessments are well validated (using historical analogues, of course) and, across models’ generations, reassessed, and the uncertainties well quantified. On the other hand, robustness in this work is implied only or mainly as the exploration of a correct set of scenarios which the authors propose.
There are two reasons why I think this is problematic: the first is that it is unclear how many of the scenarios proposed by the authors would be helpful for understanding physical processes and inter-model uncertainties. Indeed some of those scenarios essentially do not assess a world in which SRM is implemented, but rather assess a world in which SRM is not implemented or fails, and in which what is assessed is mainly a specific assumption about “moral hazard” and the impacts that changes in emissions due to failed/not efficient SRM would have on climate. On the topic of moral hazard, which the authors often mention, it is important to note that they never specifically identify what they mean by it, nor do they acknowledge existing literature on the topic. It is, instead, taken as a given, with (unreferenced) phrases like “SRM’s moral hazard, by contrast, operates through rapid, sub-decadal climate responses coupling directly to fast political dynamic”. I don’t think this or similar assertions are supported in any way in the available experimental literature, that often finds the opposite, see for instance Andrews et al. (2022), Cherry et al. (2023); whereas if the authors wanted to partly support their assertions, they could point to Abatayo et al. (2020) and Acemoglu and Rafey (2018).
The second reason why I think the authors’ assertions about the need to consider their framework of scenarios as the only way to make valid SRM assessment are problematic is that criticizing the scenarios used as “unrealistic” to affirm that climate science assessments are wrong or biased was one of the central arguments of the now withdrawn “A Critical Review of Impacts of Greenhouse Gas Emissions on the U.S. Climate” (Climate Working Group, 2025). Quoting from there “Widespread use of RCP8.5 as a no-policy baseline has created a bias towards alarm in the climate impacts literature.”; this argument is not much different from the one the authors use here. One can of course criticize single scenarios (they are, after all, not meant to be predictions!), and propose different ones, but only insofar as it remains clear that the underlying science done on top of these scenarios is valid and useful, or prove otherwise. The IPCC reports are no less robust because they use scenarios that might (or might not) be unrealistic, or currently deprecated. Why, in the case of SRM, this should be different is something the authors justify because “This simplification can be justified (for greenhouse gas mitigation modeling) by societal and techno-economic as well as physical considerations. On the societal and techno-economic side, future emission trajectories are dominated by long-term developments including population dynamics, economic development and emission intensity of the economy.” and “The climate system’s response to GHG emissions is slowly emerging over multi-decadal horizons. This lag effectively enables “fast terms” to be averaged out when modeling the climate effect of emissions pathways”. I don’t think this is persuasive, especially since there is a lot of emerging science showing that, if one takes the broader view of short-lived forcers and not just CO2, climatic impacts arising from specific emission pathways can emerge quite soon (see Samset et al., 2025 or Gettelman et al., 2024 as examples). Does the emergence of these new understandings about rapid climate responses to mitigation mean that current climate change assessments do not meet “minimum conditions for assessment” criteria? I don’t think so, but it would flow from the authors’ arguments. Similarly, there are other MIPs that, very legitimately, use idealized pathways to explore models' behavior, and whose conclusions are then used for policy-relevant decision making or for broad dissemination of results. With a large overlap in authorship list, ZECMIP is one of those (Jones et al., 2019), and now flat10MIP. I always found utterly puzzling the (widespread) criticisms of ZECMIP results about zero emission commitments because they're too idealized, or don't include some key dynamics like aerosol emissions in most simulations. The simplicity of the scenarios devised is a merit, and the inter-model spread they show an incredibly important result.
My third main point deals with the practicalities of what the authors propose. Section 4 concludes with the assertion that the authors are identifying “what must be represented” to meet their threshold of a meaningful assessment. However, section 5 starts by saying that “This paper does not prescribe a specific modeling solution”. So… what? The final assertion from the Conclusions seems to be that, if one thinks about the large unpredictability of societal and geopolitical behaviors, there is simply no way to model or conceptualize any future, and there should be no attempt to do so, because it will fail; but that, while this is not that big of a problem for climate mitigation, predictive power over societal behavior is such a fundamental part of any eventual SRM assessment that simply nothing of value could be said about what SRM would do to the climate, and thinking otherwise is “such an acute risk”. Ultimately then, as the authors say in the last phrase, the framework they themselves propose is not even an attempt at a solution. It’s just pointing at a very complicated scheme on a board and saying “see, this is just too complicated, let’s just give up”. I don’t think this is true, but mainly I don’t see this as being useful. I think current research, rather than being naive in assessing “SRM primarily through “best-case” scenarios”, is very purposely at the stage where its main service is in identifying knowledge gaps and highlighting areas of certainty and uncertainty. Plenty of research that is available right now based on “naive” scenarios is already capable of identifying and discussing trade-offs and uncertainties in the context of SRM (to avoid other references, let me just point the authors to the rather comprehensive list of research in particular emerging from the Global South here: https://www.degrees.ngo/published-research/ ), and much has been learned about models’ behavior, risky strategies (overcooling of the tropics under equatorial injections, or hemispherically-imbalanced SRM) and potential impacts across a variety of Earth systems, from ozone impacts to agriculture. How would these assessments have benefitted from the broader set of scenarios proposed here? Would they have yielded more robust insight into the inner working of climate, or the discrepancies between models? The authors need to make the case for why the specific realizations of chaos (the sub-optimal, failure-mode scenarios) they propose are a requirement for producing the generalized insights that an overarching assessment must return, and that they think a “simpler” modeling framework doesn’t provide. Introducing chaos into the scenarios one assesses is not the best way to assess the consequences of chaos, in fact, it undermines the effort. Orderly, scientifically motivated experimental design can provide the greatest insights into the consequences of well-managed, idealized SRM as well as the failure modes in chaotic versions of SRM. The climate response in these simpler, cleaner experiments that yield valuable physical insight can then be leveraged in emulators and simpler modeling frameworks to illustrate the consequences of specific instances of chaos (as is happening already), without claiming that the lack of such scenarios in the context of already existing modeling framework is a great moral and scientific fault.
This doesn’t mean that non-ideal cases can’t be explored, but rather that their exploration has to be both systematic and practical. What the authors of this piece do is situate their proposal ambiguously between a simple classification of scenarios (but that was already done for instance in Lockley et al., 2022) and the proposal to actually integrate the fast-looped dynamics of both behavioral dynamics and political ones, without ever detailing how this could be done practically. But suggestions on how to do this exist: for instance, Beckage et al. (2025) proposed the inclusion of behavioral and cognitive processes in scenario exploration of SRM using social climate models (see Beckage et al., 2022). The difference between the Beckage proposal and this one is that in the other case, the explicit intent is to explore, using modeling tools that already exist and can be developed, how our own assumptions about “moral hazard” affect scenario generation, and then classify them, rather than hand-picking “ultimate moral hazard” cases a priori.
Ultimately, the authors’ assertion that previous assessments are not robust is their own, and this being a perspective, they’re within their right to have their own opinion. But I don’t think their assertion is supported or robust itself. If the authors want to pursue the publication of this piece, and the editor agrees, then I can only hope a revised version will acknowledge the presence of others’ perspectives established over time, that it will avoid misrepresenting them, and avoid too broad claims about others’ failures while failing to suggest operable solutions themselves.
References
A.L. Abatayo,V. Bosetti,M. Casari,R. Ghidoni, & M. Tavoni, Solar geoengineering may lead to excessive cooling and high strategic uncertainty, Proc. Natl. Acad. Sci. U.S.A. 117 (24) 13393-13398, https://doi.org/10.1073/pnas.1916637117 (2020).
Acemoglu, D. and Rafey, W., 2018. Mirage on the horizon: Geoengineering and carbon taxation without commitment (No. w24411). National Bureau of Economic Research.
Andrews, T.M., Delton, A.W. and Kline, R., 2022. Anticipating moral hazard undermines climate mitigation in an experimental geoengineering game. Ecological Economics, 196, p.107421.
Beckage, B., Lacasse, K., Raimi, K.T. and Visioni, D., 2025. Models and scenarios for solar radiation modification need to include human perceptions of risk. Environmental Research: Climate, 4(2), p.023003.
Beckage B, Moore F C and Lacasse K 2022 Incorporating human behaviour into Earth system modelling Nat Hum. Behav. 6 1493–502
Brody, E., Visioni, D., Bednarz, E.M., Kravitz, B., MacMartin, D.G., Richter, J.H. and Tye, M.R., 2024. Kicking the can down the road: understanding the effects of delaying the deployment of stratospheric aerosol injection. Environmental Research: Climate, 3(3), p.035011.
Buck HJ (2022) Environmental Peacebuilding and Solar Geoengineering. Front. Clim. 4:869774. doi: 10.3389/fclim.2022.869774
Cherry, T.L., Kroll, S., McEvoy, D.M., Campoverde, D. and Moreno-Cruz, J., 2023. Climate cooperation in the shadow of solar geoengineering: an experimental investigation of the moral hazard conjecture. Environmental politics, 32(2), pp.362-370.
Climate Working Group (2025) A Critical Review of Impacts of Greenhouse Gas Emissions on the U.S. Climate. Washington DC: Department of Energy, July 23, 2025
Diao, C., Keys, P. W., Bell, C. M., Barnes, E. A., & Hurrell, J. W. (2025). A model study exploring the decision loop between unilateral stratospheric aerosol injection scenario design and Earth system simulations. Earth's Future, 13, e2024EF005455. https://doi.org/10.1029/2024EF005455
Estrada, F., Bastien-Olvera, B.A., Calderon-Bustamante, O., Altamirano, M.A., Muñoz-Sánchez, R., Moreno-Cruz, J. and Botzen, W., 2026. Economic assessment of SRM under socio-political and geophysical tipping dynamics. Environmental Research: Climate, 5(1), p.015015.
Laakso, A., Kokkola, H., Partanen, A.-I., Niemeier, U., Timmreck, C., Lehtinen, K. E. J., Hakkarainen, H., and Korhonen, H.: Radiative and climate impacts of a large volcanic eruption during stratospheric sulfur geoengineering, Atmos. Chem. Phys., 16, 305–323, https://doi.org/10.5194/acp-16-305-2016, 2016.
Lloyd, Elisabeth A. “Confirmation and Robustness of Climate Models.” Philosophy of Science 77, no. 5 (2010): 971–84. https://doi.org/10.1086/657427.
Jackson, L. S., Crook, J. A., Jarvis, A., Leedal, D., Ridgwell, A., Vaughan, N., & Forster, P. M. (2015). Assessing the controllability of Arctic sea ice extent by sulfate aerosol geoengineering. Geophysical Research Letters, 42(4), 1223–1231. https://doi.org/10.1002/2014GL062240
Jones, C. D., Frölicher, T. L., Koven, C., MacDougall, A. H., Matthews, H. D., Zickfeld, K., Rogelj, J., Tokarska, K. B., Gillett, N. P., Ilyina, T., Meinshausen, M., Mengis, N., Séférian, R., Eby, M., and Burger, F. A.: The Zero Emissions Commitment Model Intercomparison Project (ZECMIP) contribution to C4MIP: quantifying committed climate changes following zero carbon emissions, Geosci. Model Dev., 12, 4375–4385, https://doi.org/10.5194/gmd-12-4375-2019, 2019.
Keys, P. W., 2023. The plot must thicken: a call for increased attention to social surprises in scenarios of climate futures, Environ. Res. Lett. 18 081003
Kwiatkowski, L. et al 2015. Atmospheric consequences of disruption of the ocean thermocline Environ. Res. Lett. 10 034016
Farley, J., MacMartin, D.G., Visioni, D. and Kravitz, B., 2024. Emulating inconsistencies in stratospheric aerosol injection. Environmental Research: Climate, 3(3), p.035012.
Farley, J., MacMartin, D. G., Visioni, D., Kravitz, B., Bednarz, E., Duffey, A., and Henry, M.: A Climate Intervention Dynamical Emulator (CIDER) for Scenario Space Exploration, Geophysical Model Development [preprint], https://doi.org/10.5194/egusphere-2025-1830, 2026, accepted.
Gettelman, A., Christensen, M. W., Diamond, M. S., Gryspeerdt, E., Manshausen, P., Stier, P., et al. (2024). Has reducing ship emissions brought forward global warming? Geophysical Research Letters, 51, e2024GL109077. https://doi.org/10.1029/2024GL109077
Lockley, A., Xu, Y., Tilmes, S., Sugiyama, M., Rothman, D., & Hindes, A. (2022). 18 Politically relevant solar geoengineering scenarios. Socio-Environmental Systems Modelling, 4, 18127. https://doi.org/10.18174/sesmo.18127
MacMartin, D.G., Visioni, D., Kravitz, B., Richter, J.H., Felgenhauer, T., Lee, W.R., Morrow, D.R., Parson, E.A. and Sugiyama, M., 2022. Scenarios for modeling solar radiation modification. Proceedings of the National Academy of Sciences, 119(33), p.e2202230119.
McLaren, D. and Corry, O.: Clash of Geofutures and the Remaking of Planetary Order: Faultlines underlying Conflicts over Geoengineering Governance, Global Policy, 12, 20–33 2021
Pflüger, D., Wieners, C. E., van Kampenhout, L., Wijngaard, R. R., & Dijkstra, H. A. (2024). Flawed emergency intervention: Slow ocean response to abrupt stratospheric aerosol injection. Geophysical Research Letters, 51, e2023GL106132. https://doi.org/10.1029/2023GL106132
Samset, B.H., Wilcox, L.J., Allen, R.J. et al. East Asian aerosol cleanup has likely contributed to the recent acceleration in global warming. Commun Earth Environ 6, 543 (2025). https://doi.org/10.1038/s43247-025-02527-3
Trisos, C.H., Amatulli, G., Gurevitch, J., Robock, A., Xia, L. and Zambri, B., 2018. Potentially dangerous consequences for biodiversity of solar geoengineering implementation and termination. Nature Ecology & Evolution, 2(3), pp.475-482.
Quaglia, I., Visioni, D., Bednarz, E.M., MacMartin, D.G. and Kravitz, B., 2024. The potential of stratospheric aerosol injection to reduce the climatic risks of explosive volcanic eruptions. Geophysical Research Letters, 51(8), p.e2023GL107702.
Visioni, D., Kravitz, B., Robock, A., Tilmes, S., Haywood, J., Boucher, O., Lawrence, M., Irvine, P., Niemeier, U., Xia, L., Chiodo, G., Lennard, C., Watanabe, S., Moore, J. C., and Muri, H.: Opinion: The scientific and community-building roles of the Geoengineering Model Intercomparison Project (GeoMIP) – past, present, and future, Atmos. Chem. Phys., 23, 5149–5176, https://doi.org/10.5194/acp-23-5149-2023, 2023a.
Wan, J.S., Chen, CC.J., Tilmes, S. et al. Diminished efficacy of regional marine cloud brightening in a warmer world. Nat. Clim. Chang. 14, 808–814 (2024). https://doi.org/10.1038/s41558-024-02046-7
Wiertz, T. (2015). Visions of Climate Control: Solar Radiation Management in Climate Simulations: Solar Radiation Management in Climate Simulations. Science, Technology, & Human Values, 41(3), 438-460. https://doi.org/10.1177/0162243915606524 (Original work published 2016)
Visioni, D., Bednarz, E. M., Lee, W. R., Kravitz, B., Jones, A., Haywood, J. M., and MacMartin, D. G.: Climate response to off-equatorial stratospheric sulfur injections in three Earth system models – Part 1: Experimental protocols and surface changes, Atmos. Chem. Phys., 23, 663–685, https://doi.org/10.5194/acp-23-663-2023, 2023.