Original title: Optimizing Geoengineering Interventions Using Differentiable Climate Models
Pulkit Dubey, Dorian S. Abbot, Ashesh Chattopadhyay · September 2026
Geoengineering aims to cool Earth, but badly designed efforts could cause serious disruptions. Scientists need ways to figure out the best cooling strategies before anyone tries them. In this study, researchers used a computer model that simulates the atmosphere using math that allows for automatic optimization. They simulated a world where the oceans warmed by +4 K. They then tested where to apply cooling across five ocean bands to bring land temperatures back to normal. This cooling mimics real-world methods like brightening marine clouds or spraying aerosols into the stratosphere. Because weather becomes unpredictable over time, the researchers optimized the cooling in short chunks of 8 to 14 days. Their strategy removed 92.3 ± 0.4% of land warming. Surprisingly, focusing only on land temperature also restored normal land rainfall, evaporation, and humidity. Finally, the plan worked when tested in separate AI climate models without needing changes.
You are trying to cool a hot room by opening windows. But the outside temperature changes unpredictably, so a plan that worked an hour ago might now make the room hotter. In this paper, the researchers use a climate model that can calculate the effect of each cooling intervention, and they update the plan every 8-14 days to account for changing conditions.
Geoengineering could soon be deployed to cool the planet, but poorly designed interventions risk causing new problems. This work provides a method to design cooling strategies that not only reduce land warming by over 92% but also restore precipitation and humidity patterns, even though those weren't directly targeted. That means we can potentially engineer climate interventions that are both effective and less disruptive.
This study uses an idealized setup in a single climate model with simplified ocean cooling patterns. Real-world geoengineering would involve complex aerosol dynamics and global circulation effects not captured here. The strategy was only tested over two to three years, so long-term stability and unintended consequences remain unknown. The optimization is limited to five zonal bands, so finer spatial control might be needed.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Optimizing Geoengineering Interventions Using Differentiable Climate Models — Pulkit Dubey, Dorian S. Abbot, Ashesh Chattopadhyay
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.