Same Target Temperatures, Two Very Different Climates: Long Geoengineering Simulations Reveal Hidden Divergence

Stratospheric aerosol injection, or SAI, has long been framed as a deceptively simple proposition: scatter reflective sulfate particles high in the atmosphere, mimic the cooling of a large volcanic eruption, and buy humanity time while greenhouse gas emissions are brought under control. But a new set of climate model experiments suggests that the apparent simplicity hides a deeper problem. Researchers at the NSF National Center for Atmospheric Research and the University of Colorado Boulder, publishing in Earth System Dynamics, have shown that two geoengineering strategies designed to produce exactly the same global and regional temperature patterns can end up steering the Earth system into two substantially different climate states, with profoundly different consequences for the ocean circulation that helps regulate the North Atlantic.

The team, led by Walker Raymond Lee together with Simone Tilmes and Ewa M. Bednarz, ran three extraordinarily long simulations with the Community Earth System Model version 2, or CESM2. Each lasted 150 model years, far longer than the 35-to-80-year runs that dominate the geoengineering literature. The simulations were built on the idealized framework of the Geoengineering Model Intercomparison Project’s G2 experiment, in which carbon dioxide concentrations rise by one percent per year against a preindustrial background, with no other anthropogenic forcings muddying the picture. By injecting sulfur dioxide directly into the stratosphere at pre-defined latitudes roughly five kilometers above the tropopause, the researchers used feedback algorithms to hold surface temperatures near preindustrial values throughout the century and a half of simulated warming.

Two of the new experiments, called G2-SAI-3DOF and G2-SAI-hybrid, employed the now-standard three-degree-of-freedom strategy. This approach uses independent injections at 30 degrees north, 15 degrees north, 15 degrees south, and 30 degrees south to manage three temperature metrics simultaneously: global mean temperature, the temperature difference between the hemispheres, and the equator-to-pole gradient. The third experiment, G2-SAI-1DOF, used a simpler hemispherically symmetric strategy at 30 degrees north and south, managing global mean temperature alone, mirroring the recently proposed G6-1.5K-SAI experiment. The crucial twist was that the two three-degree-of-freedom runs shared identical temperature targets but differed in how their control algorithms were initialized, with the hybrid version defaulting to tropical injections that could later be shifted poleward, while the standard version prioritized southern tropical injections from the start.

The results were striking. Both three-degree-of-freedom simulations successfully maintained the same temperature targets over 150 years, yet they converged on fundamentally different injection patterns and, ultimately, different climate states. The G2-SAI-3DOF run settled into a strategy dominated by injections at 15 degrees south, the same pattern that the influential ARISE-SAI-1.5 experiment had converged to in CESM2. The hybrid run, by contrast, drifted toward injections concentrated at 30 degrees north and 30 degrees south, resembling the simpler one-degree-of-freedom strategy. Because 15 degrees south injection cools the planet relatively evenly in this model, the controller kept favoring it, while the hybrid controller’s freedom to redistribute injections allowed a different equilibrium to emerge.

The most consequential divergence involved the Atlantic Meridional Overturning Circulation, the great oceanic conveyor belt that carries warm tropical water poleward in the North Atlantic and returns cold, deep water southward. Under rising carbon dioxide alone, the AMOC weakens substantially, driven by reduced surface heat loss and freshening in the North Atlantic, which together lower surface density and suppress the deep mixing that drives overturning. In the G2-SAI-3DOF simulation, with its predominantly Southern Hemisphere injections, this decline was slowed but not stopped. In the hybrid and one-degree-of-freedom simulations, which injected far more sulfur at 30 degrees north, the AMOC decline was halted entirely, with circulation strength maintained near preindustrial levels.

This finding builds on earlier work showing that injection latitude exerts first-order control on the AMOC response. A 2025 study by Bednarz and colleagues had already demonstrated that Northern Hemisphere injections have a much stronger restorative effect on overturning strength than Southern Hemisphere ones. The new results add a feedback loop to the picture: as the AMOC weakens, it transports less heat toward the Northern Hemisphere mid and high latitudes, altering the interhemispheric temperature gradient and pushing the controller to inject more in the tropics and Southern Hemisphere. A stronger AMOC does the opposite, encouraging more Northern Hemisphere injection. The ocean circulation and the injection strategy thus reinforce each other in opposite directions depending on where the particles are released, and small initial choices in the control algorithm can tip the system toward one branch or the other.

The implications extend beyond oceanography. The two three-degree-of-freedom simulations produced different regional temperature distributions even while hitting the same targets: the southern-injection run left the Northern Hemisphere subtropics warmer and higher latitudes cooler, while the hybrid run did the reverse. Precipitation responses diverged as well. All the SAI scenarios reduced global mean precipitation more than greenhouse warming had increased it, a well-known side effect of blocking sunlight, with the G2-SAI runs offsetting roughly 175 percent of the precipitation increase under carbon dioxide forcing, compared to about 106 to 119 percent in the shorter, scenario-based experiments. Residual warming persisted over northern Asia and the Southern Ocean in all three G2-SAI simulations, and the southern-injection strategy produced stronger drying over tropical land and in the North Atlantic region.

Armed with these insights, the team revisited ARISE-SAI-1.5, one of the most widely analyzed policy-relevant geoengineering scenarios, whose datasets have informed studies of Arctic and Antarctic ice, agriculture, monsoons, and extreme weather. The original experiment placed roughly 60 percent of its sulfur at 15 degrees south. The researchers modified the controller to shift as much injection as possible to 30 degrees north and south, producing a new three-member ensemble called ARISE-hybrid that meets the same temperature targets with about 40 percent of injections at 15 degrees south, 35 percent at 30 degrees south, and 25 percent at 30 degrees north. Even over ARISE’s short 35-year window, the differences were statistically significant: the hybrid ensemble showed a stronger AMOC, deeper North Atlantic mixed layers, warmer North Atlantic and Arctic surface temperatures consistent with enhanced poleward heat transport, increased winter precipitation in that region, and a southward shift of tropical rainfall.

The authors are careful to note that neither strategy is inherently better, and that the original ARISE design was not flawed. The point is that a second solution to the same control problem exists, and that the Earth system’s response to the two solutions differs in meaningful ways. The implicit design choices made when building a feedback controller, long treated as technical details, can shape regional climate outcomes as much as the headline temperature objectives themselves. The team recommends that future experiments document not only their targets but also the priorities, injection strategies, and motivations behind each design decision.

More broadly, the work exposes a limitation of the linear assumptions underpinning most SAI strategy design, which treats the relationship between injections, aerosol optical depth, and temperature as essentially additive. Over decades to centuries, nonlinear ocean feedbacks break that assumption, meaning the same set of temperature targets can correspond to multiple substantially different planetary states. Because the ARISE experiments are too short to reveal whether the hybrid and original ensembles would eventually diverge as dramatically as the G2-SAI runs did, the authors argue that long simulations are essential for identifying slow-onset impacts like AMOC changes. For a technology whose risks and benefits are already the subject of intense public and political debate, the message is sobering: there is no single way to implement stratospheric aerosol injection, and the span of possible outcomes, even under identical temperature goals, may be wider than anyone assumed.

Subject of Research: Divergent long-term climate responses to stratospheric aerosol injection strategies designed to meet identical temperature targets

Article Title: Exploring divergent long-term stratospheric aerosol injection scenarios with the G2-SAI and ARISE-hybrid experiments

Article References: Lee, W. R., Tilmes, S., & Bednarz, E. M. (2026). Exploring divergent long-term stratospheric aerosol injection scenarios with the G2-SAI and ARISE-hybrid experiments. Earth System Dynamics, 17(4), 1117-1134. https://doi.org/10.5194/esd-17-1117-2026

Image Credits: AI Generated

DOI: 10.5194/esd-17-1117-2026

Keywords: stratospheric aerosol injection, solar geoengineering, climate intervention, AMOC, CESM2, GeoMIP, ARISE-SAI-1.5, feedback controller, ocean circulation, climate modeling, sulfur dioxide injection, Earth System Dynamics