Solar Geoengineering (Sun Reflection)
Sunlight reflection means putting tiny particles high in the sky so that a little less sunlight reaches the ground and the planet cools a bit. A big volcano did this by accident in 1991: after Mount Pinatubo erupted, the whole world was cooler for a year or two. Nobody knows how it would change rain and monsoons in each country, which particles are safe for the ozone layer, or who should be allowed to decide. It does nothing about the extra CO2 in the air and the sea, and if it were stopped suddenly the warming would come back fast. Scientists are only now planning small, careful tests.
Open in the interactive tree →Sulfate or other particles released at about 20 km would reflect a little sunlight, as Mount Pinatubo's did in 1991, when about 17 million tonnes of sulfur dioxide cooled the Earth by roughly 0.4-0.5 degrees C for about two years. Models consistently show that moderate cooling would reduce many climate impacts within months, but changes in rainfall, ozone damage and a rapid rebound if it stopped are unresolved. It would not remove CO2 or stop ocean acidification, and the IPCC's sixth assessment says it cannot be the main response. One estimate puts delivery of 5 million tonnes a year to 20-30 km at US$2-8 billion annually.
As of now
A US Government Accountability Office review (24 March 2026) found that only a few outdoor experiments have been run, among them a marine cloud brightening test in Australia, that a US experiment was cancelled over safety and ethical concerns, and that no international consensus on rules exists; the US National Academies had called in 2021 for a research programme with a code of conduct, a public registry and permits for outdoor tests. The start-up Stardust Solutions raised US$60 million in October 2025 and published a safe-research framework in April 2026, with outdoor tests scheduled for 2026. About US$200 million was spent on research through 2024 and the UK invested over £60 million in 2025; lawmakers in at least 28 US states have proposed restrictions, and Tennessee (2024) and Florida (2025) have enacted them.
What is missing
- Climate models that agree on regional rainfall and temperature changes
- A particle that reflects sunlight without harming the ozone layer
- Aircraft or balloons able to lift millions of tonnes a year to about 20 km
- Monitoring that can tell the effect from natural variation
- International rules on who may test, deploy and stop
Becomes possible once solved
- Cap the peak of warming within months while emissions fall and CO2 is removed
- Relief for heat-sensitive reefs, ice and crops in the worst-hit regions
- More time for emissions cuts and carbon removal to take effect
Open steps
- Predict Regional Effects of Sun Reflection Medium AI leverageClimate models disagree on how reflecting sunlight would change rainfall, monsoons and extremes region by region, and on how the answer depends on where and when particles are injected.
- Choose a Safe Sun-Reflecting Particle Medium AI leverageSulfate aerosols can slow ozone recovery and add acid; alternatives such as calcite, alumina or engineered particles have not been tested at scale in the stratosphere.
- Lift Sun-Reflecting Particles to 20 km Low AI leverageAircraft or balloons would have to carry millions of tonnes a year to about 20 km and keep doing it for decades; no such fleet exists.
- Detect and Attribute Sun-Reflection Effects Medium AI leverageMeasure where injected aerosol goes and separate its effect on temperature and rain from volcanoes, El Nino and other natural variation, ideally during small experiments first.
- Rules for Starting and Ending Sun Reflection Low AI leverageNo treaty covers outdoor tests or deployment, effects would cross borders, and stopping suddenly after years of cooling would let the warming snap back.
Where today's AI could help
Medium AI leverage. Learned emulators let researchers screen many injection strategies cheaply, but results can only be checked against other models until real tests exist, and the main hurdles are physical and political.
- Train fast neural emulators on a few expensive climate-model runs
- Search injection patterns that reduce regional harm with reinforcement learning
- Analyse satellite and sensor data to detect plumes and attribute their effects
Shown so far
- A 2019 paper (Schroeder de Witt and Hornigold) proposed training a neural emulator of the HadCM3 climate model so reinforcement learning could search aerosol-injection strategies. source
- At EGU 2023 Agrawal and Schroeder de Witt trained such an emulator on HadCM3 data and released a benchmark dataset and code. source
Prerequisites
- Greenhouse Effect1896The idea is to offset greenhouse warming with reflected sunlight.
- Earth Observation by Satellite1960Satellites would have to track the particles and their effect.
- Climate Models1967Only climate models can say what reflecting sunlight would do.