Transboundary effects from idealized regional geoengineering

Regional geoengineering, by reflecting sunlight over a very limited spatial domain, might be considered as a means to target specific regional impacts of climate change. One of the obvious concerns raised by such approaches is the extent to which the resulting effects would be detectable well beyond the targeted region (e.g. in neighbouring countries). A few studies have explored this question for targeted regions that are still comparatively large. We consider idealized simulations with increased ocean albedo over relatively small domains; the Gulf of Mexico (0.23% of Earth's surface) and over the Australian Great Barrier Reef (0.07%), both with negligible global radiative forcing. Applied over these very small domains, the only statistically significant non-local changes we find are some limited reduction on summer precipitation in Florida in the Gulf of Mexico case (adjacent to the targeted region). The lack of transboundary effects suggests that governance needs for such targeted interventions are quite distinct from those for more global sunlight reflection.


Introduction
While climate change is a global challenge, the impacts are ultimately local; these include, for example, risks of increased strength of tropical storms due to increased sea surface temperatures, or risks to coral reefs due to a combination of increased ocean temperatures and acidification.There has been considerable research into 'global' solar geoengineering interventions (e.g.National Academies of Sciences, Engineering,and Medicine 2021), most frequently using stratospheric aerosol injection (SAI).
The combination of aerosol lifetime and stratospheric circulation means that SAI would yield an inherently global strategy; even 'polar-focused' approaches, (Lee et al 2023), would lead to at least hemispheric cooling.In contrast, Marine Cloud Brightening (MCB) (Latham 1990, Rasch et al 2009, Latham et al 2012a), while frequently similarly discussed as a possible tool for global cooling (e.g.Kravitz et al 2018), could be applied over much more limited spatial domains to address regional warming.The potential benefits from local interventions have already led to active research into using MCB to cool the Great Barrier Reef, for example (Tollefson 2021).MCB would involve spraying sea salt aerosols into marine stratocumulus to increase cloud reflectivity; by leveraging cloud-aerosol effects a large radiative effect can potentially be achieved with relatively modest added sea salt mass (Wood 2021).MCB would only be effective in regions with the right type of cloud conditions, but it would also be possible to exert a direct radiative effect with sea salt aerosols (Ahlm et al 2017); referred to as Marine Sky Brightening (MSB).While this would be significantly less efficient than MCB, making it impractical for large-scale cooling, MSB could provide a means to reflect sunlight over small spatial regions where MCB would not be effective due to the lack of receptive cloud types.Other means of altering surface albedo have also been proposed (Gabriel et al 2017).There are also other means of reducing sea surface temperature regionally (Ricke et al 2021) that may have similar climatic effects.
Governance is generally presumed to be easier for regional deployment than global (e.g.MacCracken 2016), but the governance challenges will be strongly affected by the extent of transboundary influence.This potential for local intervention thus raises the question of what the climate effects of a small spatial-scale intervention might be, and in particular whether the impacts would be largely confined to the country choosing to implement them (Quaas et al 2016).The answer to that question is of course dependent on the specific region, the spatial scale, and the magnitude of the desired cooling, but herein we attempt to provide some initial evidence through two concrete examples.
There have been several studies to date of regional geoengineering, often (as here) idealized.The G4foam experiment of the Geoengineering Model Intercomparison Project simulation increased ocean surface albedo over 15% of Earths area (Gabriel et al 2017) from 0.06 to 0.15, with spatially heterogeneous but broadly distributed global effects.MCB has been considered over several different spatial extents by Latham et al (2012bLatham et al ( , 2013Latham et al ( , 2014) ) over regions as small as 3% of Earth's surface area (5% of the ocean); these also lead to statistically significant changes in temperature well beyond the targeted areas.With a study question similar to ours, Dipu et al (2021) consider a cooling implemented in their climate model by altering cloud optical depth (via changed liquid water path) over North America, covering only 0.5% of Earths surface, with −9.8 ± 5Wm −2 local radiative forcing.While some small changes showed up well outside of the target region, it is unclear from their study whether these are teleconnections or simply the result of variability (as one would expect roughly 5% of the planet to appear to be statistically significant at a 95% confidence level simply due to random chance); this suggests that it is plausible that small-scale (relative to the globe) interventions might not have broader ramifications.The answer to that will clearly depend on the specific region being targeted and whether it excites teleconnections.For example, Ricke et al (2021) consider a reduction in sea surface temperature (achieved through heat flux, representative of using ocean pipes for upwelling of deeper, colder water) over the Indian Ocean, covering 5.7% of Earths surface, with the region deliberately chosen to (successfully) excite a teleconnection to reverse Sahelian droughts.There have also been several studies of regional albedo modifications over the Arctic; Cvijanovic et al (2015) for example consider idealized changes in ocean albedo over the Arctic from 65-90°N (4.7% of Earth surface) to 80-90 °N (0.76%); even in the smallest case teleconnections lead to detectable influences beyond the Arctic.Other approaches for cooling the Arctic However, all of these simulations have been conducted over regions that, while relatively small compared to the Earths surface area, are still considerably larger than local interventions that might be used to target specific local impacts.Herein we consider two case studies: the Gulf of Mexico, relevant to coral reefs and hurricane strength, and the Great Barrier Reef, also relevant to coral reefs, and a site of active MCB research.The area covered is 0.23% and 0.07% of the planet in our two examples.We use the Community Earth System Model (CESM) to simulate an intervention using an idealized modification of surface albedo (intended to be roughly representative of what might be achieved with MSB, for example).In a companion study, Goddard et al (2022) focus on the physics of injecting sea salt aerosols over the Gulf of Mexico and the resulting forcing due to direct aerosol and cloud modification effects, but the regional Weather Research and Forecasting model (WRF) used therein is not suited to exploring transboundary impacts.In contrast, the idealized simulations here clearly do not capture the physical details, but rather explore the question of, if the surface could be cooled, would there be any detectable transboundary effects (in the presence of natural variability).Section 2 describes the model and simulations, section 3 looks at the impact of cooling over the Gulf of Mexico (GM), while section 4 looks at cooling over the Great Barrier Reef (GBR).In the former case we find some limited influence on precipitation over Florida and up the US Atlantic coast, but no detectable influence at distances greater than the spatial scale of the intervention itself.For the latter we find no detectable influence outside the target region.

Model and simulations
Simulations (MacMartin et al 2023) are conducted in the fully-coupled Community Earth System Model, version 2.1 (Danabasoglu et al 2020), with a resolution of 0.95°in latitude by 1.25°in longitude.
For both the Gulf of Mexico and Great Barrier Reef, we apply a 15% increase in surface albedo for incoming shortwave radiation (visible and near-IR, and both for direct and indirect light).This is of course an idealization as injecting salt aerosols will have a different effect due to the effects on clouds (Zhao et al 2021, Goddard et al 2022).This also constitutes a relatively substantial local intervention intended to elicit a sufficient signal to noise ratio, and may not be realistic in practice, though we note that Goddard et al (2022) do achieve close to a 15% change in top-of-atmosphere shortwave forcing locally through added salt aerosols.The regions are outlined in figures 1 and 2 respectively, and correspond to an area of 1.17 × 10 6 km 2 , or 0.23% of Earth's surface, and 3.5 × 10 5 km 2 , or 0.07% of Earth's surface.
We use the preindustrial control simulation conducted as part of CMIP6.We conduct two simulations, one for each of the two regions, with each branched from the preindustrial control but with increased ocean surface albedo over the specified region.Both perturbed simulations are conducted for 51 years, with the last 50 years compared to 50 years of the preindustrial (the initial transient is short).
Averaged over the forcing region, the change in net surface shortwave (SW) in the two regions is 26 and 24 Wm −2 respectively while the change in net shortwave radiation at the top of atmosphere is 24 and 20 Wm −2 .While locally large, the global forcing is less than 0.06 Wm −2 .
To evaluate statistical significance in the results that follow, we adjust the degrees of freedom to account for serial autocorrelation in time series assuming an AR(1) process (Wilks 2019).This is not critical for precipitation, where the interannual autocorrelation is small, but is important to include in analysis of temperature changes.

Gulf of Mexico
Targeted cooling of near-surface ocean temperatures over the Gulf of Mexico might reduce the intensity of hurricanes (Latham et al 2012b) and effects of heat stress on coral reef ecosystems (Dee et al 2019).Our goal here is not to assess the feasibility of achieving local radiative forcing (explored in Goddard et al 2022) nor efficacy in cooling and subsequent effects on these impacts, but simply to look at the detectability of non-local effects.
Increasing the albedo over the Gulf of Mexico in our simulations results in local cooling of about 0.5 °C in annual-mean sea surface temperature (SST) and 0.3 °C in surface air temperature, as shown in figure 1. Summer (June-July-August, JJA) cooling is slightly larger at 0.65 °C and 0.4 °C (Supplementary S3).Subsurface ocean temperatures are also reduced (figure S5).To the extent that the goal of the intervention would be a reduction in heat stress for coral reefs or a reduction in SST anomalies that would affect the strength of tropical storms passing over the region, then this (fairly large) intervention has the desired effect.A 1 °C change in global mean temperature in this climate model leads to roughly 0.7 °C change in both annual-mean and JJA SST over the Gulf (figure S9); this reduction is thus sufficient to offset the local annual-mean sea-surface warming from a roughly 0.75 °C change in global mean temperature, and the summer warming from a nearly 1 °C change in global mean temperature.A smaller (in magnitude, spatial extent or seasonal extent) intervention may still be significant in reducing maximum summer temperatures.The reduction in SST extends into the Florida Strait (consistent with the direction of ocean currents).
The reduction in incoming SW absorbed and SST results in decreased evaporation from the Gulf (figure S1), and decreased precipitation over the Gulf (figure 1(d)).The surrounding ocean provides the moisture source for precipitation over Florida.Decreased moisture transport off the Gulf thus leads to a statistically significant reduction in precipitation over Florida of 0.3 mm/day (9% reduction), relative to natural variability of where the change relative to preindustrial is statistically significant at a 95% level using 2-sided t-test, with the degrees of freedom adjusted to account for autocorrelation.0.45 mm/day; more than half of the reduction occurs in the summer (figures S3, S4).While the details of smallscale convective precipitation over Florida are not resolved at the grid-scale of the model the physical mechanism underlying the change is plausible-reducing the source of moisture can be expected to lead to a reduction in precipitation.The effect on precipitation extends somewhat further up the Atlantic coast as well as shown in figure 1.
Despite the significant forcing level, and including 50 years of simulations for detecting changes, there are no regions with statistically significant changes that are further away from the targeted region than the spatial scale of the targeted region.While there is slightly cooler water flowing out of the Gulf through the Florida Strait, there is no statistically significant effect on the strength of the Atlantic Meridional Overturning Circulation (figure S6; strength is 22.5 ± 0.2 in preindustrial control, and 22.6 ± 0.2 with intervention, where the range indicates standard error) or SST in the North Atlantic.There is considerable variability in AMOC at multidecadal timescales and so there could be an effect that is simply not detectable over our 50-year simulations.Nor is there any statistically significant teleconnection into the Pacific Ocean.Maps of the global response are shown in figure S7; for any variable there are of course some grid cells that appear to have a statistically significant response at a 95% confidence, but the total area of these is less than the 5% one would expect from random chance.

Great Barrier Reef
Corals in the Great Barrier Reef and elsewhere are under severe stress from climate change (Hughes et al 2017).This has motivated consideration of using MCB, among other methods, to cool the ocean surface (Tollefson 2021), with preliminary trials ongoing.While MCB may not prove to be effective here, given local cloud conditions, over this small a geographic region it is plausible that MSB may prove effective even though it is less efficient.In our simulations, increasing surface albedo by 15% leads to a 22 Wm −2 reduction in surface shortwave forcing averaged over the targeted region, and 20 Wm −2 SW reduction at top-of-atmosphere.The change in sea surface temperature over the targeted region is statistically significant, but small, at 0.25 °C.In CESM2, each 1 °C increase in global mean temperature leads to roughly a 0.7 °C increase in SST over the Great Barrier Reef (figure, S9); this cooling is thus equivalent to offsetting the change in SST over the reef from a 0.35 °C change in global mean temperature.Achieving a more substantial reduction in temperature would require even larger forcing over the target region or the forcing to be applied over a much larger area.Changes in evaporation are statistically significant, but changes in surface air temperature and precipitation are not statistically significant even over the target region; this is true both for the annual-mean response and in any specific season.Unsurprisingly, there are no statistically significant changes in areas remote from the target region (beyond isolated grid points that make up much less than 5% of the surface area and are thus more likely a result of natural variability); see figure S8 and the stipplings in the south-west corner of the maps in figure 2(a), (d).
A more regional model may be more appropriate to quantifying the relationship between the magnitude of forcing and the expected cooling over the reef (and in particular the reduction in extremes).However, in addition to highlighting the lack of transboundary implications from such an effort, these results further suggest that achieving significant reduction in impacts from local interventions may be challenging.

Discussion
Much of the discussion surrounding the concept of sunlight reflection methods (SRM; known variously as climate engineering, solar geoengineering, solar radiation modification, or solar climate intervention) considers a global deployment aimed at countering a range of climate change impacts globally.It is well recognized that approaches such as Marine Cloud Brightening could be applied locally to address specific impacts, and generally assumed that the governance challenges would be simpler for more local-scale deployments.This relies on the assumption that the influence from a local implementation would also be local or at most regional.
There have been a few studies of moderate scale geoengineering, over areas from 3%-5% of the Earths surface, but very few at much smaller scales relevant for local impacts; an exception is Dipu et al (2021) who consider forcing over continental North America covering roughly 0.5% of the Earths surface.Here we explore the potential effects of much more local deployments that might target local impacts: the Gulf of Mexico (area of 1.17 × 10 6 km 2 , covering only 0.23% of the planet), and the Great Barrier Reef (area of 3.5 × 10 5 km 2 covering only 0.07% of the planet).These regions were selected specifically because of the potential to mediate climate change risks associated with hurricanes (Latham et al 2012b) and coral reefs (Latham et al 2013, Dee et al 2019, Tollefson 2021); the latter is already a site of active research.In the Gulf of Mexico case we find a reduction in precipitation over Florida, adjacent to the targeted region.However, in neither case do we find any statistically significant influence well outside of the targeted region beyond isolated grid points (figures S7, S8) that are most likely from random chance.This is despite conducting 50-year simulations with fairly significant forcing (15% increase in ocean albedo).This suggests that indeed local deployments could be amenable to local or regional governance, and further that the combination of multiple such efforts would still likely not have a global effect (in contrast to concerns implicitly expressed through suggested restrictions on simultaneous experiments, for example (National Academies of Sciences, Engineering,and Medicine 2021)), though of course the specifics will depend on the specific regions targeted and in particular whether those are deliberately chosen to excite teleconnections (Ricke et al 2021).
The simulations are idealized (by altering ocean surface albedo) rather than directly simulating a specific physical mechanism.Simulations are also conducted only in a single climate model.If some actor wanted to deploy regional geoengineering, further research would be appropriate to address both of these limitations.Further reductions in non-local effects might also be obtained by, for example, limiting the intervention only to summer (if it is the reduction in maximum summer temperature that is desired).
have similarly been shown to have remote influences (Nalam et al 2018, Sun et al 2020, Lee et al 2023).

Figure 1 .
Figure 1.Changes in surface shortwave forcing (top left), sea surface temperature (top right), surface air temperature (bottom left), and precipitation (bottom right), for reduction in surface albedo over Gulf of Mexico (region outlined in green).Averaged over the target region, the annual-mean change in shortwave radiation is 26 Wm −2 and 24 Wm −2 at the surface and top of atmosphere respectively; sea surface temperature is reduced by 0.5 °C, surface air temperature by 0.3 °C, and precipitation by 0.3 mm/day.Additionally, precipitation averaged over Florida (region outlined in red) is also reduced by 0.3 mm/day.Stippling indicates grid cellswhere the change relative to preindustrial is statistically significant at a 95% level using 2-sided t-test, with the degrees of freedom adjusted to account for autocorrelation.

Figure 2 .
Figure 2. As in figure 1, but for forcing over the Great Barrier Reef.Averaged over target area, change in surface and top-of-atmosphere shortwave forcing are statistically significant (average reduction 22 and 20 Wm −2 reduction respectively), and sea surface temperatures are reduced by 0.25 °C.Note that the colour scale is chosen to highlight the lack of statistically-significant effects; nonlocal changes are still small compared to natural variability and not statistically-significantly different from zero.