Stratospheric injections from explosive volcanic eruptions are routinely incorporated in many global climate models used for future climate projections. Therefore, the majority of SAI studies model the emission of sulfur dioxide into the stratosphere, but on a continuous basis rather than the pulse-like injection characteristic of volcanic eruptions. Studies of the cooling potential of stratospheric sulfur dioxide suggests that the strongest radiative forcing occurs for emissions in the mid-stratosphere (21–24 km) (e.g., Jones et al., 2017; Marshall et al., 2019). Some exploration of other sulfur-based materials exists using direct aerosol injections of H₂SO₄ (Pierce et al., 2010; Vattioni et al., 2019; Janssens et al., 2020; Weisenstein et al., 2022) and carbonyl sulphide (COS, Quaglia et al., 2022), but the benefits in terms of resulting cooling efficiency depend on the details of the injection strategy and have been contested in the case of using COS (e.g., von Hobe et al., 2023). Materials other than sulfate aerosol have also been considered with the aim of reducing stratospheric heating or impacts on ozone (e.g., Ferraro et al., 2011; Jones et al., 2016; Dykema et al., 2016; Keith et al., 2016; Vattioni et al., 2023a). However, physical and chemical properties, including details of coating with other aerosols in the stratosphere, are largely unknown (Dai et al., 2020; Vattioni et al., 2023a), owing to the scarcity of detailed laboratory studies at stratospheric conditions and the lack of suitable analogues (Haywood J. et al., 2022).

Modelling studies have shown that, for an appreciable cooling impact, SAI delivery systems need to be capable of delivering millions of tonnes of aerosols or their gaseous precursors to the stratosphere at altitudes of around 20 km each year (e.g., Robock et al., 2009). Currently, relatively few studies explore the feasibility and costs of an SAI delivery system (Robock et al., 2009; Smith and Wagner, 2018; Smith, 2020; Smith et al., 2022a; de Vries et al., 2020; Janssens et al., 2020). Such studies highlight the lack of feasibility for an immediate implementation but a relatively low cost and high feasibility for implementation in the next one to two decades compared to the costs of adaptation or other climate change mitigation measures. Costs are such that any deployment could feasibly be implemented by a number of governments or entities acting on their behalf. Studies also underscore the potential trade-offs between the benefits of higher injection altitudes and the complications of reaching those altitudes (Janssens et al., 2020; Smith et al., 2022b). Research into delivery mechanisms is contentious as it could be conceived as being the “slippery-slope” to deployment but the technical feasibility issues are critical in informing deployment scenarios for climate modelling purposes.

Analyses aimed at understanding the trade-offs and economic costs of any SAI deployment need to go beyond the simple expense of delivering the material. Preliminary efforts quantifying costs in an Integrated Assessment Modelling framework (IAMs, Moreno-Cruz et al., 2012; Belaia et al., 2021) found challenges in properly identifying a comprehensive damage function (i.e., the integration of an appropriate basket of metrics that best represent the detrimental impacts of climate change as a function of temperature). Such challenges included considering framing beyond temperature-dependent ones, monitoring costs, differentiating impacts depending on deployment strategies, regional trade-offs (Clark et al., 2023), residual impacts, risk of failure (Quaas et al., 2017), and potential compensation mechanisms between winners and losers. Such issues are not just confined to SRM research as there are similar concerns over CDR and conventional mitigation. Furthermore, as IAMs do not include a human-behaviour dimension, it might be hard to capture all the potential interactions between societies (Beckage et al., 2022), international politics and governance (Reynolds and Horton, 2020) over the full range of potential SAI deployment scenarios and strategies.

Major research gaps related to generation and delivery of SAI include:

The lack of quantification of technical barriers and costs. There is very limited work on understanding of technological limitations of different delivery systems.

Many of the key aerosol processes, i.e., nucleation, condensation, and coagulation occur at a microphysical scale and therefore climate models need to represent the sub-grid scale processes via physical or statistical parameterisations. This can be done by representing the aerosol population in a variety of ways ranging from simple to complex. The simplest schemes transport aerosol mass and assume a fixed size distribution for the whole population; this method is computationally cheap relative to more sophisticated methods. More complex modal schemes assume that the aerosol size distribution can be represented by three or more overlapping log-normal distribution modes, and that an aerosol can evolve from one mode to another (e.g., Mann et al., 2010). The most detailed, and most computationally expensive sectional schemes do not assume any a-priori size distribution but represent aerosols of different sizes in multiple bins and allow particles to evolve between bins (e.g., Tilmes et al., 2023). Simulating aerosol size distributions with fidelity is of fundamental importance in aerosol-radiation interactions, owing to the sensitivity of reflection of sunlight to the radius of the particles; maximal impact occurs when the radius of the aerosols is similar to the wavelength of incident sunlight (Mie, 1908; Dykema et al., 2016). The surface area density of aerosol size distributions also influences heterogeneous chemical processes. Furthermore, aerosol sedimentation velocity and the lifetime of aerosol depend on the details of the aerosol size distribution. Only a few models exist with fully interactive stratospheric chemistry, 3-D atmospheric dynamics and aerosol microphysics (Kleinschmitt et al., 2017, 2018; Tilmes et al., 2022), and even the most advanced models struggle to reproduce aerosol size distributions after volcanic events (Quaglia et al., 2023). Process studies (Vattioni et al., 2023b) and observations of moderate volcanic eruptions (e.g., Wrana et al., 2023; Li et al., 2023) have shown strong inter-eruption dependency of the resulting stratospheric aerosol size distributions, with some unexpected decreases in the size of aerosols after some eruptions (Wrana et al., 2023). Lack of understanding of the processes and their inter-dependency result in significant uncertainties in the atmospheric burden for a specified injection rate (Weisenstein et al., 2022).

Further limitations when modelled aerosol size distributions are compared against those from volcanic eruptions arise from the pulse-like nature of the injection, while SAI simulations generally either maintain or increase emission rates over many years (e.g., Kravitz et al., 2013). Stratospheric burdens in many SAI simulations are therefore frequently many times larger than the well observed eruptions that have occurred in the satellite record. Theoretically, these enhanced burdens and continuous emission rates should lead to increases in aerosol sizes owing to enhanced condensation and coagulation rates (e.g., Niemeier and Timmreck, 2015), but supporting observational evidence is lacking. Additional limitations of using volcanic eruptions as natural analogues for SAI include the fact that other gaseous species (e.g., H₂S, HCl, Cl, Br, water vapour, e.g., Vömel et al., 2022) and other aerosols (e.g., ash, Kloss et al., 2021; Wells et al., 2023; sea salt, Colombier et al., 2023) are frequently co-emitted with SO₂ which influences the observed aerosol size distributions.

While satellite observations are very useful for observing the evolution of stratospheric SO₂ to sulfate aerosol at large spatial scales, retrieval limitations exist in accurately determining the evolution of aerosol size distributions. Furthermore, satellites can only provide limited information on the initial behaviour of injected plumes owing to saturation effects coupled with coarse temporal sampling, which frequently means that plumes are only observed once or twice a day. Limited outdoor experimentation and/or comprehensive laboratory studies under stratospheric conditions could potentially provide such information (Golja et al., 2021; Vattioni et al., 2023a) and could be extended to solid aerosols (such as diamond, TiO₂, calcite, or alumina) and account for their interaction with naturally occurring stratospheric aerosols (Dai et al., 2020).

Aerosol size distributions are intrinsically linked to ozone depletion as they provide a surface area on which heterogeneous halogen-induced chemical reactions occur. Absorption of solar and terrestrial radiation by stratospheric aerosols directly affects stratospheric temperatures, which can change stratospheric dynamics and influence the transport of ozone-rich air from the tropics towards polar regions (e.g., Haywood J. et al., 2022). Estimates of ozone depletion over Antarctica in spring (i.e., when the ozone hole is at its peak) could reach approximately 60 ± 20 DU for SAI deployments sufficient to achieve a 0.5°C global cooling over the period 2020–2040. Over the Arctic, the corresponding total column ozone depletion (in Dobson Units, DU) is uncertain, having been estimated as 13 DU ± 10 DU and 22 DU ± 21 DU by two models (Haywood J. et al., 2022). For reference, at the height of the ozone hole (around the year 2000), Antarctic and Arctic total column ozone decreased from unperturbed values of around 300 DU and 450 DU, respectively, to less than 150 DU and 375 DU, respectively. Some increases in total column ozone in mid-latitudes owing to changes in the dynamical transport occur, but these are very uncertain and very model dependent.

The net effect of SAI on stratospheric ozone depends on the availability of ozone-depleting substances present in the stratosphere for heterogeneous chemical reactions (e.g., Xia et al., 2017). Because ozone depleting substances have been effectively controlled by the Montreal Protocol (e.g., Laube et al., 2022), their concentrations are projected to decline in the future. Thus, SAI is modelled to have less impact on stratospheric ozone the later any deployment occurs (Tilmes et al., 2021; Tilmes et al., 2022; Haywood J. et al., 2022), but this area has not been well investigated owing to the limited number of SAI scenarios and strategies examined to date. Changes in ozone and aerosol concentrations ultimately affect the quantity and quality of surface radiation, especially at ultraviolet wavelengths.

Finally, SAI produces aerosols that are transferred into the troposphere through sedimentation and tropopause folds where they could act as either cloud condensation nuclei or ice nuclei. These impacts have been assessed in a few studies but have not been comprehensively investigated (Section 5.1).

Major Research Gaps related to the process-level understanding of SAI include:

The lack of model sophistication. Only a few models represent the complex gas-phase and heterogeneous chemistry processes, aerosol microphysics, and atmospheric dynamics of sulfate-based SAI with fidelity.

Volcanic eruptions reveal that to achieve a long-lasting perturbation to aerosol optical depths and hence a considerable cooling of climate, aerosols or their gaseous precursors need to be injected at altitudes of around 20 km at the tropics, but lower altitudes at higher latitudes (e.g., Jones et al., 2017; Marshall et al., 2019). Pulse injections from large explosive volcanic eruptions in the tropics can result in aerosol plumes that spread into both hemispheres (e.g., Pinatubo in 1991; e.g. McCormick et al., 1995), into primarily the northern (El Chichon in 1982; Hofmann, 1987), or primarily the southern hemisphere (Agung in 1963; Rampino and Self, 1982) depending on the prevailing dynamical conditions within the stratosphere, and the altitude and latitude of the eruption.

Recognising that differences in climate model results depend on the scenario, deployment strategy, process-level treatment of aerosol microphysics, and differences in the dynamical response of global climate models, GeoMIP was established (Kravitz et al., 2013) which aimed to standardise modelling efforts across multiple climate models. Much of the focus of GeoMIP has been on SAI, although early idealised GeoMIP simulations also included the impacts of idealised reductions in the solar constant and idealised MCB simulations. Within GeoMIP, multiple intercomparisons have been carried out by a range of climate models, also across different Coupled Model Intercomparison Project (CMIP) iterations, and almost 150 papers have been published based on its results (Visioni et al., 2023b). A further multi-model SAI experiment has been coordinated by the Community Climate Modelling Initiative (CCMI; Plummer et al., 2021).

Certain robust residual climate impacts were identified for deployment strategies that are limited to the equator, including an over-cooling of tropical regions and continued warming at polar regions when compared to a world warmed solely by greenhouse gases with the same global mean temperature. However, risk–risk analyses (i.e., the climate response of a global warming world to those of a world where SAI is used to abate climate warming) reveal that these regions are far less impacted than under global warming, and most associated regional climate extremes are mitigated under SAI (e.g., Curry et al., 2014; Jones et al., 2018).

To reduce any residual climate impacts, more recent studies proposed injection strategies that use multiple sulfur injection locations, to adjust the resulting aerosol latitudinal distribution and the resulting stratospheric aerosol optical depth (Kravitz et al., 2016, 2017; Tilmes et al., 2018; Visioni et al., 2023a; Henry et al., 2023). A control-feedback loop can be used to adjust the magnitude of emissions at multiple locations (Kravitz et al., 2016, 2017) to achieve a number of surface climate targets, with the intent of reducing residual impacts from climate change (Kravitz et al., 2019; Tilmes et al., 2018; Richter et al., 2022; Henry et al., 2023). These climate targets typically include global mean temperature, inter-hemispheric temperatures, and equator-pole temperature gradients. While nominally identical targets can be achieved, the resulting injection strategies (in terms of the latitudinal distribution and magnitude of injections) show considerable inter-model variation (Henry et al., 2023) and dependency on the modelled scenarios (Wells et al., 2024); the optimal latitudinal distribution for any SAI deployment in the real world is therefore unknown. Other studies include regional and global precipitation and sea-ice change targets (e.g., Lee et al., 2020) emphasising significant limitations as to how many climate targets can be controlled simultaneously.

The most comprehensive multi-model assessments of the efficacy of the radiative forcing, near-surface cooling, and injection rates required to cool the Earth by 1°C are summarised in Haywood J. et al. (2022) (Table 1).

The large uncertainties surrounding the metrics that are diagnosed by the different climate models are due to differences in model representations of stratospheric chemistry, transport, radiation and aerosol microphysical processes. However, the near surface cooling is broadly consistent with those subsequent to the eruption of Mount Pinatubo which injected a pulse of around 10–15 Tg SO₂ and cooled the planet by around 0.5°C.

Major Research Gaps related to SAI scale and deployment strategies include:

The large uncertainties in metrics associated with SAI. These limit assessments of the effectiveness of SAI as a climate intervention technique in reaching climate targets.

SAI would affect surface climate patterns through two main pathways: (i) reducing sunlight, thereby cooling the surface and (ii) locally affecting temperatures in the stratosphere, thereby affecting stratospheric circulation (e.g., Visioni et al., 2021; Haywood J. et al., 2022). Furthermore, the sensitivity of global-mean precipitation per unit temperature change differs for GHGs and SAI; a unit warming from GHGs results in less increase in precipitation than the decrease in precipitation for a unit cooling from SAI (e.g., Irvine et al., 2019; Visioni et al., 2023a,b). This feature is well understood, having been investigated in a number of studies (Bala et al., 2008; Niemeier et al., 2013; Modak and Bala, 2014; Modak et al., 2016). Additionally, the spatial distributions of the warming from GHGs and the cooling from SAI differ. Thus, reflecting sunlight via SAI cannot perfectly cancel out the CO₂ warming signal, and while it would result in lower overall temperatures, it would not perfectly restore the climate. Multi-model simulations indicate that such residual warming patterns from SAI cooling would be minor compared to any large unabated warming signal from GHGs (e.g., Tilmes et al., 2018; Jones et al., 2018; Visioni et al., 2021).

SAI induces a large-scale reduction in precipitation and soil moisture (Tilmes et al., 2013; Cheng et al., 2019), that also is discernible in observations following the eruption of Pinatubo (Trenberth and Dai, 2007). Observations and model simulations suggest that precipitation patterns may be significantly affected in some regions under certain SAI scenarios and strategies (Haywood et al., 2013; Irvine et al., 2019; Jones et al., 2022). For instance, the location of the intertropical convergence zone (ITCZ), which regulates tropical precipitation, may be influenced depending on inter-hemispheric differences in cooling/warming, also affecting monsoonal precipitation (Oman et al., 2006; Haywood et al., 2013; Da-Allada et al., 2020; Krishnamohan and Bala, 2022). Other precipitation patterns may be affected under significant SAI deployments owing to changes in the position and strength of the storm tracks in mid-latitudes (e.g., Jones et al., 2021, 2022) that depend strongly upon the latitudinal dependence of the injection strategy that influence the equator-pole temperature gradients (e.g., Lutsko et al., 2020; Bednarz et al., 2023; Visioni et al., 2023b), or changes in circulation in tropical oceans (Pomalegni et al., 2022). These impacts stem from a combination of uneven cooling, changes in stratospheric circulation, and stratospheric-tropospheric coupling.

The Quasi-Biennial Oscillation, which is evident in observations as a reversal of equatorial stratospheric winds with a period of around 2 years has been shown to be strongly influenced or eradicated by the absorption of solar and terrestrial radiation by sulfate aerosols (e.g., Aquila et al., 2014; Jones et al., 2016; Jones et al., 2022; Haywood J. M. et al., 2022), but the consequences in terms of surface impacts have received little attention, and the impacts are lessened if SAI strategies avoid injecting at equatorial latitudes (e.g., Kravitz et al., 2019; Bednarz et al., 2023; Wells et al., 2024).

Major Research Gaps related to the large-scale circulation response to SAI include:

The lack of characterization of the stratospheric circulation. The representation of the stratospheric circulation differs significantly between models and in model-observational analysis.

Together with an expanded climate modelling effort, there is a need to consider regional impacts far beyond average climatic variables. This includes ecosystems (managed and unmanaged land-based ecosystems and marine ones) that may be affected, for instance, by changes in climatic extremes, incoming solar radiation and sulfur deposition (Trisos et al., 2018; Zarnetske et al., 2021). Similarly, studies that explore the climate-health (Eastham et al., 2018; Carlson et al., 2022) and climate-food (Fan et al., 2021; Clark et al., 2023) nexus in the context of SRM are also fundamental to properly inform expected impacts for different populations, and for meaningful engagement with impacted communities. Currently, there is a paucity of such studies, which would require both increased involvement from the related ecology and epidemiology communities and a strengthening of the tools necessary for assessing impacts in these areas. This could be achieved by improving the dialogue between physical climate modellers and ecology and epidemiology specialists, by sharing data, and co-design of scenarios.

Major Research Gaps related to the impacts of SAI include:

Lack of understanding of ecosystem response. There is a general lack of studies on how single species and entire ecosystems respond to concurrent changes in CO₂ concentrations, temperature, precipitation and quality of light.

The most prominent proposed generation mechanism of additional CCN is by spraying sea water through modified nozzles of commercial sprayers used in agriculture or in industry for dust-suppression. However, commercial sprayers used in dust control generate droplets of typically ~100 micron size, which is an order of magnitude larger than cloud droplets and two orders of magnitude larger than the sea water droplet size required for the resultant aerosol to act as efficient CCN. Various techniques for spraying small droplets are summarised in Cooper et al. (2013) and Connolly et al. (2014). Alternate methods to sprays such as wet grinding in an organic solute, or cryochemical methods are available for making sea-salt aerosol of a suitable size (Zhilenko et al., 2016), but the scale required remains a formidable obstruction. Use of sea water also represents a challenge as the nozzles would need to be flushed very regularly to remove build-up of sea-salt and impurities. Owing to the location of low-lying marine clouds over open oceans, the most prominent proposed delivery system is via a dedicated large fleet of ships.

Commercial sprayers that are currently available operate at flow rates of up to around 100 litres per minute (lpm), equivalent to an injection of dry sea-salt aerosol (NaCl) of around 3.5 kg per minute which equates to around 1800 tonnes/year if run for 24 h per day for 365 days of the year. Production of sea-salt at an optimal mono-disperse size is a significant challenge owing to rapid coalescence (Cooper et al., 2013) leading to the presence of a small number of giant CCN (GCCN), which can initiate precipitation leading to the loss of cloud water and break up stratocumulus clouds (e.g., Feingold et al., 1999; Pringle et al., 2012; Jensen and Nugent, 2017; Dziekan et al., 2021; Hoffmann and Feingold, 2021). Evaporative cooling (Jenkins and Forster, 2013) can cause the plume to subside, preventing the generated aerosol from reaching cloud-base. The high flow rates of 100 lpm when producing sub-micron sea-salt aerosols also cannot currently be achieved using modified nozzles on commercially available sprayers where rates are typically a few litres per minute.

Global modelling studies suggest that there is considerable uncertainty in the optimal size of the generated sea-salt aerosol, meaning that the technologically challenging goals highlighted above are unclear. The majority of global climate model simulations of aerosol-cloud interactions rely on a Köhler-theory-based aerosol activation scheme (i.e., representation of the conversion of aerosol particles into cloud droplets) of Abdul-Razzak and Ghan (2000). However, limitations of this activation scheme have been highlighted particularly for the marine aerosols typically suggested for MCB. Nenes and Seinfeld (2003), Ming et al. (2006), and Morales Betancourt and Nenes (2014) have developed alternative activation schemes for global models, which have been implemented in some Earth System models. Climate model studies using these different activation schemes suggest that injection of aerosols generally increases CDNC, but reductions can occur if even a small number of giant aerosols (>250 nm dry radius) are injected into clouds with low cloud-base updraft velocities (Pringle et al., 2012). More recent systematic investigations have shown that cloud droplet number concentrations strongly depend on the size of the sea-salt aerosol injected into clouds. The climate model study by Haywood et al. (2023) suggested that sea-salt aerosols with a dry radius of around 90 nm were most effective at increasing cloud droplet number concentrations when using the Abdul-Razzak and Ghan (2000) activation scheme. However, process-level modelling studies (e.g., Connolly et al., 2014; Wood, 2021) suggest aerosols of 15–50 nm dry radius appear optimal. Thus, on a global mean basis, there are considerable uncertainties that propagate through the choice of activation scheme and the optimum size distribution for sea-salt aerosol that impact the relationship between mass of sea-salt emitted and the resulting cloud droplet number concentration. These uncertainties will be even greater on a regional or cloud-regime basis.

While the issues noted above undoubtedly present serious challenges, other aspects may enhance the efficiency of delivery. Continuous operation is unlikely to provide the most effective delivery as marine stratocumulus clouds typically thin during the day when exposed to sunlight and thicken overnight in the absence of solar heating. High resolution modelling studies have probed the impacts of injecting aerosols into different cloud-precipitation regimes (Wang et al., 2011) including injections at different times of day (Jenkins et al., 2013; Prabhakaran et al., 2024). These studies are consistent in suggesting that results appear strongly dependent upon whether the baseline clouds are precipitating or non-precipitating and reveal that weakly precipitating unpolluted clouds seeded in the morning exert the strongest radiative forcing through either reduction in loss of LWP (Jenkins et al., 2013) or enhanced cloud fraction (Wang et al., 2011; Prabhakaran et al., 2024). Thus, optimal strategies might focus on delivery in the morning in areas of weakly precipitating clouds. Further efficiencies could be envisaged through the injection of aerosol higher in the marine boundary layer where losses through deposition to the surface are reduced. However, neither high-resolution nor GCM simulations have robustly quantified these impacts.

Major Research Gaps related to MCB generation and delivery include:

The lack of consistency between aerosol activation schemes. Different schemes give different results and they may be pushed beyond their design limits in modelling extensive MCB deployments.

The impact of emissions of sulfate aerosols from sulfur-rich marine fuels on ship-tracks has received considerable attention for some time (e.g., Capaldo et al., 1999; Lauer et al., 2007; Eyring et al., 2010; Partanen et al., 2012; Christensen et al., 2022). Ship-tracks and larger-scale volcanic eruptions provide plentiful evidence of reductions in cloud droplet effective radius (e.g., Toll et al., 2017, 2019; Christensen et al., 2022) when aerosol concentrations are enhanced. While assessing the fidelity of GCM responses from individual ship-tracks is difficult (e.g., Glassmeier et al., 2021), larger scale perturbations from effusive volcanic eruptions or in larger scale shipping lanes (Diamond et al., 2020) provide more relevant comparisons. Simulations of the reduction in cloud droplet effective radius by multiple models of the large-scale effusive eruption of Holuhran (Malavelle et al., 2017; Jordan et al., 2024) reveal that GCM scale models are able to capture the magnitude and the spatial distribution of the aerosol-induced perturbation to the cloud effective radius. Generally, the mean from the models performs better than the individual models, a feature that has been noted in several studies for other climate model variables (e.g., Gleckler et al., 2008). However, evidence of cloud liquid water changes under polluted conditions are inconsistent, with observational studies frequently diagnosing weak cloud liquid water decreases (e.g., Malavelle et al., 2017; Gryspeerdt et al., 2019; Toll et al., 2017, 2019). While some global climate models replicate this weak impact on cloud liquid water with reasonable fidelity, others do not (e.g., Malavelle et al., 2017).

There is new observational evidence from effusive volcanic eruptions (e.g., Chen et al., 2022) and ship-tracks (Manshausen et al., 2022) that the cloud-fraction response to increased concentrations is more significant than previously diagnosed. The response of cloud fraction to increases in aerosol concentrations appears weaker in GCMs than in observations. Further evidence that the local cloud adjustments do not significantly offset the expected microphysical brightening of clouds comes from the reduction in sulfur content in marine fuel from 3.5 to 0.5% in 2020. This follows International Maritime Organisation (IMO) regulations designed to reduce pollution and provides a natural experiment to examine aerosol-cloud interactions over remote ocean regions. Yuan et al. (2022) report an almost 50% reduction in visible ship-track occurrence contributing a radiative forcing estimated to be as strong as approximately +0.3 W m⁻² although March et al. (2021) point out that some reductions in ship-track occurrence may be due to the concomitant reductions in shipping owing to the impacts of Covid-19. Watson-Parris et al. (2022) account for Covid-19 reductions and find a 25% reduction in ship-tracks for an estimated 80% reduction in sulfur emissions indicating that even in remote shipping lanes, pristine conditions are not always present. Diamond (2023) estimate a radiative forcing due to adoption of IMO regulations of around 1Wm⁻² in major shipping corridors suggesting that the global estimates from climate models of approximately +0.1Wm⁻² (+0.07 to +0.15Wm⁻², Gettelman et al., 2024) appear reasonable. Climate model studies that assess the impacts of the IMO shipping regulations on global mean temperatures (e.g., Jordan and Henry, 2024; Gettelman et al., 2024) suggest that IMO regulations will have brought forward global warming by 3–4 years, but definitive attribution is difficult owing to rapid changes in anthropogenic aerosol emissions, particularly in Asia, and owing to the concurrent El Nino.

Taken together, these results suggest that there is a significant aerosol-induced microphysical brightening of clouds (Twomey, 1991) that appears to be enhanced by increases in cloud cover. However, evidence of any enhancement (or abatement) in cloud liquid water path appears contradictory, which means that the conventional sequential chain of interactions highlighted by Haywood and Boucher (2000) may be incorrect.

Major Research Gaps related to MCB process-level understanding include:

Lack of representation of key microphysical processes. Aerosol activation, cloud updraft, entrainment, and turbulence are all sub-gridscale in GCMs and are therefore heavily parameterised.

The most efficient clouds to target with MCB are low-lying, unpolluted stratocumulus clouds which exert a cooling of climate as they strongly increase the local albedo (see Equation 1), have little compensating impact on terrestrial radiation (e.g., Slingo, 1990) and are susceptible to MCB injections as evident from ship-tracks. Satellite-based relationships of cloud susceptibility have been derived (e.g., Quaas et al., 2009; Quaas et al., 2008), but have been questioned owing to the lack of sensitivity of satellite retrievals at low aerosol concentrations (Ghan et al., 2016; Ma et al., 2018).

Early climate modelling studies of the potential impacts of MCB (e.g., Rasch et al., 2009; Jones et al., 2009) simply increased the reflectance of low-lying marine stratocumulus clouds by setting cloud droplet number concentration (CDNC) to an asymptotic maximum that was informed by aircraft observations (e.g., Martin et al., 1994; Jones et al., 2001). These early studies were subsequently improved upon by more explicit modelling through the injection of sea-salt aerosol (Jones and Haywood, 2012; Partanen et al., 2012). For example, Jones and Haywood (2012) explicitly modelled sea-salt injection rates dependent on wind speed according to Korhonen et al. (2010) and targeted the most susceptible 10% of cloudy areas. Owing to difficulties in distinguishing the climatic response from the differences due to the specific climate intervention scenario or strategy, coordinated GeoMIP simulations were performed (e.g., Alterskjaer et al., 2013; Kravitz et al., 2013; Ahlm et al., 2017; Stjern et al., 2018).

The earliest GeoMIP study relevant to MCB was the G3-SSCE experiment (Alterskjaer et al., 2013) where the top-of-atmosphere radiative forcing was maintained at 2020 levels in a scenario with rising greenhouse-gas concentrations. The three participating models treated sea-salt with different degrees of complexity ranging from fully prognostic sea-salt and CDNC, through using a climatology of sea-salt concentrations and diagnostic CDNC, to prescribed sea-salt and CDNC. The results show very different cloud responses and susceptibilities, but common findings include a suppression of latent heat flux and precipitation over the low-latitude oceans, and shifts in the hydrological cycle associated with an enhanced Walker circulation. This change to the Walker circulation appears confined to MCB strategies as simulations of the same global mean cooling scenario with either SAI or space mirrors reveal little impact (Niemeier et al., 2013). Subsequently, a simpler GeoMIP experiment was defined (G4cdnc; Kravitz et al., 2013) where a 50% increase in the CDNC of low marine clouds was imposed over the oceans on a global basis; the simplicity of this experimental design meant that nine climate models were able to participate (Stjern et al., 2018) and confirmed strengthening of the Walker circulation for MCB deployment. A more complex GeoMIP experiment called G4sea-salt (Kravitz et al., 2013) was performed by three models that could all explicitly represent sea-salt injection into the marine boundary layer at latitudes between 30°S-30°N; this experiment highlighted that the aerosol direct effect could contribute a significant fraction of the modelled cooling through so-called “Marine Sky Brightening” (MSB, Ahlm et al., 2017; Jones and Haywood, 2012). The finding that MSB could be an effective method is corroborated by recent work that modelled sea-salt emissions targeting brightening of the susceptible stratocumulus clouds of the eastern Pacific (Haywood et al., 2023).

Rasch et al. (2024) analyse the results from UKESM1, E3SM, and CESM2 climate models and suggest a protocol for MCB-style simulations where injections are confined to specific oceanic regions. However, the findings suggest a great degree of variability in the radiative forcing per unit sea-salt injection, with values of around −13, −23, and −140 mW Tg⁻¹ yr. for UKESM1, E3SM, and CESM2 accordingly. This factor of ten discrepancy translates to a large uncertainty in the temperature response per unit sea-salt injection; according to simulations with the UKESM1 earth system model (Haywood et al., 2023), the maximum cooling efficiency for sea-salt emissions is around −20 mK Tg⁻¹ yr. Simulations with the CESM2 model using the same Abdul-Razzak and Ghan (2000) activation scheme suggest a maximum cooling efficiency of around −200 mK Tg⁻¹ yr. (Rasch et al., 2024). Jones and Haywood (2012) diagnosed an MCB-induced cooling of 0.54 K if the most susceptible 10% of marine clouds were targeted using a seeding size distribution similar to that of naturally occurring sea-salt, while Haywood et al. (2023) suggest an MCB-induced cooling of up to 1 K might be achievable for their specific deployment strategy targeting the eastern Pacific with MSB-induced cooling being responsible for coolings exceeding this. The MSB efficiency derived in the UKESM1 model is about 4 times smaller than for MCB at around −4 mK Tg⁻¹ yr. (Haywood et al., 2023).

Marine cloud brightening could conceivably be used, not to target the global mean temperature, but to ameliorate some of the symptoms of global warming. For example, Latham et al. (2014) suggest that MCB strategies could target maintenance of the Arctic sea-ice. Latham et al. (2014) propose that this could be achieved not through the general cooling of the Earth’s climate to reduce the flow of energy from the tropics to the poles in the atmospheric and oceanic circulations, but through a targeted high latitude MCB deployment. While this strategy is undoubtedly complicated by the presence of ice clouds (Kravitz et al., 2014), observational evidence from a large-scale effusive volcanic eruption in Iceland (e.g., McCoy and Hartmann, 2015; Malavelle et al., 2017; Chen et al., 2022) show a definitive decrease in the cloud droplet effective radius over vast swathes of the northern Atlantic, and machine learning algorithms have diagnosed an aerosol-induced increase in cloud fraction over northern latitudes (Chen et al., 2022). Climate model studies (Zoëga et al., 2024) show a more complex Arctic temperature response to this effusive eruption with a high latitude cooling during the summer when insolation levels are high, but a high latitude warming in winter owing to the limited insolation combined with an enhanced trapping of outgoing terrestrial radiation from increased cloud cover and cloud optical depth.

Observations of cumulus-dominated regions have also been shown to be sensitive to increased aerosol concentrations via large-scale effusive eruptions on Hawaii (e.g., Eguchi et al., 2011; Mace and Abernathy, 2016; Malavelle et al., 2017; Chen Y. et al., 2024) indicating that cloud regimes dominated by clouds other than stratocumulus are also susceptible to increased concentrations of aerosols suggesting that MCB could potentially be effective on a more global basis or in targeting local cooling. This observational evidence combined with the finding that marine sky brightening could potentially contribute a significant cooling suggests that marine environments of sufficient scale exist for an appreciable cooling through the emission of sea-salt into the boundary layer. A further area of investigation is whether MCB strategies could conceivably be used to induce a localised cooling over ecologically sensitive regions such as over the Great Barrier Reef off the east coast of Australia (Harrison, 2024). Indeed, vessels equipped with state-of-the-art spray technologies have recently been deployed in this area in a limited scientific outdoor investigation of the feasibility of this technique.

The magnitude of the deployment needed to determine a particular level of cooling is currently unknown. However, given that a recent state-of-the-art climate model study (Haywood et al., 2023) suggested that around 50 million tonnes of optimally-tuned dry sea-salt would need to be injected into optimally selected susceptible clouds to cool the climate by 1°C gives some idea of the current challenges in scaling up the technology for MCB. Around 27,000 sprayers working at 100 lpm, 24 h a day for 365 days per year producing optimally sized aerosols, deployed to optimally susceptible clouds would be required.

Major Research Gaps related to MCB scale and deployment strategies include:

Only very idealised MCB deployments have been simulated. Deployment strategies that target susceptible clouds in a more intelligent and progressive manner within GCMs are required.

Rasch et al. (2024) showed that the pattern of inter-model responses in temperature and precipitation to perturbing specific geographical regions using the same effective radiative forcing is remarkably similar. This result suggests that common responses exist between climate models and presents a pathway for designing deployment strategies that minimise any detrimental residual climate responses.

Modelling studies have targeted specific susceptible cloud regions (Rasch et al., 2009; Jones et al., 2009; Jones and Haywood, 2012; Partanen et al., 2012; Alterskjaer et al., 2013; Kravitz et al., 2013; Ahlm et al., 2017; Stjern et al., 2018; Hirasawa et al., 2023; Haywood et al., 2023). An aerosol lifetime of a few days to a few weeks is typical within the troposphere meaning that the cooling associated with MCB deployments that target specific regional areas is frequently localised and very inhomogeneous. If globally significant cooling is to be achieved, very much stronger localised cooling is frequently evident in the region of deployment.

Impact assessments of MCB are sparse compared to those of SAI. Should MCB be applied over the South East Atlantic region, then detrimental responses to such regionalised applications of MCB have been noted (e.g., Jones et al., 2009; Jones and Haywood, 2012; Hirasawa et al., 2023) with reductions in precipitation over the Nordeste and Amazonian regions of Brazil, a response that appears robust across models (Rasch et al., 2024). Idealised simulations (Jones et al., 2009) and explicit sea-salt injection simulations (Jones and Haywood, 2012) yield similar responses, suggesting that sea-surface temperature (SST) changes are the primary driver of changes in the Walker circulation that influence precipitation. These conclusions are supported by observed robust correlations between highly reflective clouds over the south-east Atlantic, the associated localised SST reduction, and rainfall over the Nordeste region of Brazil (e.g., Hastenrath, 1990; Utida et al., 2019).

Detrimental regional impacts are much reduced should deployments be applied at much larger scales. For example, the three models that applied MCB over the region 30⁰N-30⁰S did not find significant reductions in precipitation compared to GHG-induced impacts over any land areas (Alterskjaer et al., 2013). The multi-model simulations of Ahlm et al. (2017) and Stjern et al. (2018) show a significant, and potentially beneficial, increase in precipitation over countries surrounding the Mediterranean Sea and over Australia. The similarity of these results appears to be due to the similarity in the large-scale forcing patterns that are applied in each study.

The detrimental impact of MCB over the South East Atlantic on precipitation over South America has led some studies to focus solely on susceptible clouds over the east Pacific (Haywood et al., 2023). Depending on quantity of sea-salt emitted, this deployment strategy has the potential to induce undesirable side-effects caused by a La Niña-like response many times stronger than natural variability. The impacts of the subtropical gyre ocean circulations are highlighted as playing a key role in redistributing the MCB-induced SST anomalies. Evidence for a La Niña-like climate response produced by this and similar injection strategies (more cooling in the eastern compared to the western Pacific), is also found in Rasch et al. (2009), Jones and Haywood (2012), Hill and Ming (2012), Hirasawa et al. (2023), and Chen C. C. et al. (2024). Detrimental impacts from a permanent strong La Niña-like response include shifts in tropical and extratropical precipitation and are likely to exert other far-reaching impacts. For example, Haywood et al. (2023) show that sea-level rise over the low-lying islands of the south Pacific under their deployment scenario and strategy is greater than in the unmitigated greenhouse-gas warming scenario. Hirasawa et al. (2023) suggest that the strong increase in precipitation over Australasia that was also found in Haywood et al. (2023) originates from forcing the southeast Pacific and that, of the three regions investigated in their study, this region produces the strongest global cooling response.

Even though the number of studies is quite limited, it is clear that strong regional forcing patterns have the potential to induce strong regional responses. The more uniform forcing patterns deployed in some modelling studies, and in SAI deployments, suggests a more uniform global response.

Major Research Gaps related to large-scale circulation response to MCB include:

Model dynamical responses of regional cooling and associated inter-model consistency need to be thoroughly investigated using a suitable risk–risk framework.

The impacts of MCB on terrestrial ecosystems are intrinsically linked to the impacts on large scale dynamics which influence precipitation, temperature, and net primary productivity and are therefore not differentiated in Section 4.4. However, the impacts of MCB on marine ecosystems are almost entirely absent in the scientific literature. The lack of certainty in the effectiveness of MCB couples with the lack of a specific deployment scenario, lack of a specific deployment strategy, and uncertainties in subsequent large-scale dynamical impacts in the atmosphere and ocean resulting in too large a parameter space to comprehensively assess this issue.

If MCB were deployed, it is likely that there may be both local and remote impacts on marine ecosystems. These impacts will depend strongly on the deployment scenario and strategy. It is likely that local impacts would include those induced by reduced downwelling solar radiation at the surface, strong local reductions in SST, and changes in surface winds and oceanic currents. As many studies have indicated a La Niña-like response following the preferential targeting of susceptible clouds of the eastern Pacific, impacts might be expected to follow those evident in naturally occurring La Niña events. Positive impact on the fishing industry of western South America owing to oceanic upwelling at the west of the Pacific bringing nutrient-rich waters to the surface might be expected. However, such an assumption is dangerous as permanent La Niña conditions some five times greater than that of naturally occurring La Niñas could occur under a deployment that reduces the global mean temperature from that of SSP5-8.5 to that of SSP2-4.5 (Haywood et al., 2023). This unprecedented state would very likely bring unintended and unpredictable consequences and could potentially lead to large-scale marine ecosystem impacts.

Major Research Gaps related to impacts of MCB include:

A lack of engagement with the marine biology/ecological community. As MCB scenarios and strategies are still relatively under-developed, there has been little engagement between the physical scientists engaged with GCM modelling and the marine biology/ecological community.

SRM through CCT was first proposed by Mitchell and Finnegan (2009), and has since been investigated in several global modelling experiments (e.g., Storelvmo et al., 2013; Muri et al., 2014; Gasparini and Lohmann, 2016). These modelling experiments have differed widely in approach, with the simplest ones merely increasing ice crystal fall speed in order to mimic the result of actual INP injection (e.g., Muri et al., 2014; Kristjánsson et al., 2015; Jackson et al., 2016). These studies have consistently found the proposed cooling through increased LW infrared emission to space to occur. Despite being highly idealised, they allowed for investigations of the climate response to a proxy for actual INP injection. The validity of using such a proxy for studies of CCT climate response has been tested and confirmed (Gasparini et al., 2017). Consistent findings across these types of studies were (i) considerable SW compensation for the desired LW cooling effect, which makes CCT most effective under large solar zenith angle conditions, (ii) an enhancement of the hydrological cycle compared to a pre-industrial control climate when CO₂ warming is cancelled by CCT (for mechanism, see Kristjánsson et al., 2015), and thus potential to avoid the suppression of precipitation found in comparable multi-model SRM studies targeting incoming solar radiation, and (iii) amplified cooling of the Arctic, more closely mirroring the geographic patterns of greenhouse gas induced warming than equatorial injections of SAI, which tend to lead to undercooling of polar regions. A GeoMIP experiment (“G7cirrus”) is currently being carried out in order to explore these effects more deeply based on a consistent set of experiments across multiple models (Kravitz et al., 2015).

However, an underlying assumption in the above studies is that INP injection does in fact have the intended outcome of producing larger ice crystals with higher fall speeds. Global modelling studies that have attempted to explicitly simulate the cirrus microphysical changes induced by INP increases have found a number of challenges to this assumption. These include (i) large uncertainty about what proportion of present-day cirrus clouds form through homogeneous nucleation (a prerequisite for CCT to work; Storelvmo and Herger, 2014), (ii) the risk of under-seeding or overseeding (Storelvmo et al., 2013), and (iii) the possibility of inducing cirrus formation under certain conditions that would otherwise not be conducive to cirrus formation (Gasparini and Lohmann, 2016). These are all factors that could either lead to negligible net impact on cirrus clouds and thus no cooling effect, or worse, lead to the opposite of the desired effect.

Despite these potential pitfalls, coordinated experiments with two global climate models that had displayed large differences in previous CCT studies owing to different scenarios and strategies led to several robust conclusions: (i) CCT, when optimised in terms of seeding amounts, could according to these experiments lead to a global ERF between −1.8 and −0.8 Wm⁻², which in these two models equates to offsetting between 30 and 70% of the warming from CO₂ increases experienced to date and (ii) CCT could potentially reduce estimated climate damages caused by the CO₂ increase experienced since pre-industrial by between 50 and 85% (Gasparini et al., 2020). These simulations assumed extensive intervention in the form of globally uniform cirrus seeding. However, to optimise cooling, CCT should in fact not be globally uniform, but rather target high-latitude cirrus clouds in the winter hemisphere only (Storelvmo and Herger, 2014; Liu and Shi, 2018). The fact that all explicit simulations of INP injections have so far been carried out with only two GCMs (ECHAM-HAM and CESM) is concerning, as one model consistently finds a relatively weak or negligible CCT forcing (e.g., Tully et al., 2022; Gasparini et al., 2020), while the other generally reports CCT to be a potent SRM method (e.g., Storelvmo and Herger, 2014; Shi et al., 2024). Coordinated multi-model CCT experiments of explicit INP injection are urgently needed to fully assess the feasibility of CCT.

Cirrus clouds may also inadvertently be affected by SAI (Visioni et al., 2020), mainly through temperature changes due to the sulfate aerosol heating affecting temperature and relative humidity in the upper troposphere, but also in this respect uncertainties remain large. In a global modelling study, Kuebbeler et al. (2012) also found cirrus thinning in response to SAI due to warming and stabilisation of the upper troposphere. This thinning contributed 60% to the overall cooling due to SAI. A study explicitly considering the injection of calcite into the stratosphere (see Section 3.2) likewise found significant but poorly constrained cirrus responses to the stratospheric aerosol injection (Cziczo et al., 2019).

Compared to CCT, far less research has been conducted on MCT so far. However, from the limited studies available, it appears that some of the concerns associated with CCT are not applicable for MCT. That being said, MCT may have its own problems and pitfalls that have yet to be identified because so few studies have been conducted. The first study to investigate MCT, a regional cloud-resolving modelling experiment initially designed to investigate Arctic CCT, was an accidental one (Gruber et al., 2019). The study conducted high-resolution regional modelling of the Arctic, and was among the first to actually simulate the injection of aerosols acting as INPs in the Arctic upper troposphere (as opposed to assuming some background concentration of seeding INPs), and therefore could also simulate the response of the lower-altitude Arctic mixed-phase clouds as the injected particles sedimented down through the troposphere. The unintended effect on Arctic mixed-phase clouds amounted to between 4 and 5% reduction of their cloud cover in these simulations, and was as expected accompanied by strong additional LW infrared cooling.

The only study to date that intentionally targeted MCT was conducted by Villanueva et al. (2022). In a global climate model, they injected INPs (or in some experiments increased droplet freezing rates) in high-latitude mixed-phase clouds, and found significant high-latitude cooling given the right timing and INP dosage (approximately −1 K over the Arctic Ocean, and −0.6 K over the Southern Ocean). The cooling was accompanied by a recovery of annual mean Antarctic and Arctic sea ice, but with a very uneven seasonal distribution exhibiting a strong winter recovery partly compensated by spring/summer rebound, most likely associated with increased poleward heat transport. The study found no risk of overseeding (in contrast to CCT), but also limited scalability, as MCT is spatially and temporally limited to winter hemisphere high-latitude mixed-phase clouds. As these findings all rely on a single modelling study, the above conclusions cannot yet be considered robust.

Based on recent research, it now appears likely that a clear separation of CCT and MCT is not possible in practice. Seeding of the upper troposphere with the intention of targeting cirrus would almost certainly also impact mixed-phase clouds at lower altitudes. Likewise, but to a lesser extent, some of the seeding material that would be intended for low/mid-level mixed-phase clouds would likely make it to higher levels in the atmosphere and impact cirrus. There is also the possibility of SAI-related solid particles such as calcite acting as INP when re-entering the troposphere through tropopause folds (Cziczo et al., 2019).

While there are significant unknowns related to the optimal seeding material and strategy as well as associated environmental impacts for CCT and MCT, addressing these is premature until more fundamental research gaps related to cirrus and mixed-phase cloud thinning have been addressed.

Major Research Gaps related to CCT and MCT include:

SRM approaches that aim to modify surface albedo can be divided into land-based and ocean-based surface albedo modifications. The former have been studied more extensively to date, primarily with the use of global climate models. For example, Irvine et al. (2011) implemented three types of land surface albedo modification, targeting urban, desert and cropland areas. The global coverage of these areas at the time amounted to an estimated 0.56, 1.8 and 3.1%, respectively. For all three regions, surface albedo enhancement at the upper end of the realistic ranges led to only minor effects on global temperature, but did result in regional cooling, albeit with strong seasonality. In all three cases, and especially for the desert region, the albedo enhancement resulted in a significant reduction in rainfall over land.

Despite the issues identified above, it has been argued that regional surface albedo modification should be considered more prominently as part of regional climate adaptation and mitigation efforts (Seneviratne et al., 2018). This argument is rooted in the finding that surface albedo modifications in agricultural and densely populated areas could significantly alleviate hot extremes in these regions. Furthermore, it is argued that regional surface albedo modification may potentially avoid some of the issues identified in SAI or large-scale MCB. A GeoMIP test bed experiment (Kravitz et al., 2019) systematically exploring land surface albedo modification has been proposed for potential adoption by GeoMIP as an official core experiment (named “Land-GeoMIP”), and could reveal whether the findings of Seneviratne et al. (2018) are robust across multiple models.

Ocean surface albedo enhancement, which could be achieved through the generation of foams or microbubbles in the ocean surface, has been explored for example for Arctic Ocean surface areas in global modelling experiments, as a means to avoid sea ice melt and associated surface-albedo feedback mechanisms (Mengis et al., 2016). In a set of experiments with different emission scenarios for greenhouse gases and other climate forcers, the ocean albedo was set to 0.8 (the approximate albedo of sea ice) whenever sea ice cover dropped below 50% in model grid cells poleward of 70°N. While sea ice trends or permafrost thaw could not be reversed in any of these experiments, sea ice loss was delayed by between 25 and 60 years depending on the scenario considered, and permafrost thaw was likewise significantly delayed. Kravitz et al. (2018) performed analysis of idealised multi-model simulations performed under GeoMIP whereby ocean albedo was increased to balance an instantaneous quadrupling of carbon dioxide concentrations and compared the results to similar simulations that reduced the solar constant. Many similarities were noted in the comparison including residual warming over polar regions.

Targeting lower latitudes instead, Gabriel et al. (2017) conducted a similar experiment, but modified the surface albedo of the major subtropical ocean gyres in the Southern Hemisphere. The hemispheric asymmetry was deliberate owing to the potential desirability of a shift in the rain-bearing ITCZ to enhance precipitation over sub-Saharan Africa (e.g., Haywood et al., 2013). The ocean gyres specifically were selected because they are largely cloud free, thus increasing the efficacy of the surface albedo enhancement, and because they have low wind speeds, weak currents and limited biological productivity. In simulations following the RCP6.0 emission pathway (Meinshausen et al., 2011), the surface albedo of the SH ocean gyres were set to 0.15 (from an unperturbed value of around 0.06), and the associated climate response was evaluated. The albedo enhancement led to a global cooling of approximately 0.6 K, accompanied by increased precipitation over tropical land areas, many of which are at present under severe water scarcity threats. The experiment, referred to as “G4foam,” was a so-called GeoMIP test bed experiment proposed for potential adoption by GeoMIP as an official core experiment, but has so far only been conducted by a single model, and no formal experiment protocol has been developed to further this approach. Major hurdles associated with the viability of ocean surface albedo modification exist at present, for example regarding the feasibility of creating a stable, highly reflective layer of microbubbles at the ocean surface, and largely unexplored but likely significant effects on ocean circulation and marine biology and chemistry.

Major Research Gaps related to surface albedo modification include:

Land-based surface albedo modification: the effective radiative forcing and/or cooling that could be achieved through albedo modifications of the land surface is poorly constrained, and robust features of the associated climate have yet to be identified.