Quan et al. (2025) Solar Geoengineering Strategies Based on Reinforcement Learning

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Journal of Geophysical Research Atmospheres
• Year: 2025
• Date: 2025-12-04
• Authors: Heng Quan, Daniel D. B. Koll, Nicholas J. Lutsko, Janni Yuval
• DOI: 10.1029/2025jd044319

Research Groups

Not specified in the abstract.

Short Summary

This paper investigates the use of reinforcement learning (RL) to optimize stratospheric aerosol injection (SAI) strategies within an idealized global climate model, demonstrating that RL can learn stable, plausible, and time-dependent deployment strategies to maximize benefits and minimize side-effects.

Objective

• To investigate whether reinforcement learning (RL) can effectively control the distribution of stratospheric aerosol concentration in an idealized global climate model (GCM) to optimize solar geoengineering strategies, aiming to maximize benefits while minimizing undesirable side-effects such as shifts in rainfall patterns.

Study Configuration

• Spatial Scale: Global (within an idealized GCM), focusing on the stratosphere.
• Temporal Scale: Multiple GCM simulations, exploring the impact of geoengineering initiation time.

Methodology and Data

• Models used: Idealized Global Climate Model (GCM), simple energy-balance model.
• Data sources: Simulated data generated by the idealized GCM, used for reinforcement learning training.

Main Results

• Reinforcement learning (RL) successfully learns to produce stable and plausible stratospheric aerosol injection (SAI) strategies within several dozen GCM simulations.
• The optimal geoengineering strategy is found to be dependent on the time when geoengineering is initiated.
• This time-dependency of the optimal strategy is further explained using a simple energy-balance model.
• The study provides a first proof-of-concept that RL can identify promising SAI strategies.

Contributions

• Presents the first proof-of-concept for applying reinforcement learning to optimize stratospheric aerosol injection strategies.
• Introduces a novel feedback-control approach for solar geoengineering optimization, moving beyond previous linear algorithms.
• Demonstrates that optimal geoengineering strategies can be time-dependent, a finding explained by an energy-balance model.

Funding

Not specified in the abstract.

Citation

@article{Quan2025Solar,
  author = {Quan, Heng and Koll, Daniel D. B. and Lutsko, Nicholas J. and Yuval, Janni},
  title = {Solar Geoengineering Strategies Based on Reinforcement Learning},
  journal = {Journal of Geophysical Research Atmospheres},
  year = {2025},
  doi = {10.1029/2025jd044319},
  url = {https://doi.org/10.1029/2025jd044319}
}