Reinforcement Learning for Optimizing Climate Change Interventions: A DQN-Based Approach
DOI:
https://doi.org/10.32350.umt-air.61.05Keywords:
Reinforcement Learning, Deep Q-Network, CMIP6 SSP1-2.6, Climate InterventionsAbstract
This research presents a reinforcement learning framework based on the Deep Q-Network (DQN) and Proximal Policy Optimization (PPO) to optimize climate intervention policies under the CMIP6 SSP1-2.6 scenario. Global climate variables—near-surface temperature anomaly (relative to 2015), vertical velocity (wap), and precipitation (pr)—were processed using area-weighted averaging to avoid latitudinal bias. Interventions (carbon capture, reforestation, and inaction) apply cumulative effects to simulate persistent mitigation. After 50 episodes, DQN’s cumulative reward of -3004.26 and PPO’s -3001.98 approximate an 83% improvement over the fixed carbon-capture baseline (-17936.92) with significance (DQN vs. Baseline: t = 48.15, p < 0.0001). There was no statistical difference found between DQN and PPO (p = 0.8335). The policy that is to be learned reduces the cumulative deviation of temperature anomaly from 1.5°C target to ~1462 (higher in the baseline), the policy distribution favors inaction with the percentage of (86.69%) due to the cost, followed by reforestation (6.63%), and post-policy carbon capture (6.68%). Training took 13.33 minutes, showcasing high computational efficiency compared to GCMs. We show that RL can complement the standard routine evaluation of climate policies with an alternative but distinctive tool.
Downloads
References
[1] J.-Y. Lee et al., “Future global climate: Scenario-based projections and near-term information,” in Climate Change 2021: The Physical Science Basis: Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, V. Masson-Delmotte et al., Eds. Cambridge, UK: Cambridge Univ. Press, 2021, pp. 553–672, doi: https://doi.org/10.1017/9781009157896.006.
[2] T. P. Barnett and M. E. Schlesinger, “Detecting changes in global climate induced by greenhouse gases,” J. Geophys. Res. Atmos., vol. 92, no. D12, pp. 14772–14780, Dec. 1987, doi: https://doi.org/10.1029/JD092i D12p14772.
[3] F. Giorgi, “Climate change hot-spots,” Geophys. Res. Lett., vol. 33, no. 8, Art. no. L08707, Apr. 2006, doi: https:// doi.org/10.1029/2006GL025734.
[4] S. I. Seneviratne et al., “Changes in climate extremes and their impacts on the natural physical environment,” in Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation, C. B. Field et al., Eds. Cambridge, UK: Cambridge Univ. Press, 2012, pp. 109–230, doi: https://doi.org/10.1017/CBO9781139177245.006.
[5] [REMOVE OR REPLACE—The supplied title, authors, and DOI could not be verified as one publication.]
[6] [REMOVE OR REPLACE—The supplied title, authors, and DOI could not be verified as one publication.]
[7] K. Hasselmann, “Multi-pattern fingerprint method for detection and attribution of climate change,” Clim. Dyn., vol. 13, no. 9, pp. 601–611, 1997, doi: https://doi.org/10.1007/ s003820050185.
[8] M. Rummukainen, “State-of-the-art with regional climate models,” WIREs Clim. Change, vol. 1, no. 1, pp. 82–96, Jan.–Feb. 2010, doi: https://doi.org/10. 1002/wcc.8.
[9] P. Bauer, A. Thorpe, and G. Brunet, “The quiet revolution of numerical weather prediction,” Nature, vol. 525, no. 7567, pp. 47–55, Sep. 2015, doi: https://doi.org/10.1038/nature14956.
[10] V. Eyring, S. Bony, G. A. Meehl, C. A. Senior, B. Stevens, R. J. Stouffer, and K. E. Taylor, “Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization,” Geosci. Model Dev., vol. 9, no. 5, pp. 1937–1958, May 2016, doi: https:// doi.org/10.5194/gmd-9-1937-2016.
[11] K. Feng, N. Lin, R. E. Kopp, S. Xian, and M. Oppenheimer, “Reinforcement learning-based adaptive strategies for climate change adaptation: An application for coastal flood risk management,” Proc. Natl. Acad. Sci. U.S.A., vol. 122, no. 12, Art. no. e2402826122, Mar. 2025, doi: https:// doi.org/10.1073/pnas.2402826122.
[12] K. E. Taylor, R. J. Stouffer, and G. A. Meehl, “An overview of CMIP5 and the experiment design,” Bull. Amer. Meteorol. Soc., vol. 93, no. 4, pp. 485–498, Apr. 2012, doi: https://doi. org/10.1175/BAMS-D-11-00094.1.
[13] D. Rolnick et al., “Tackling climate change with machine learning,” ACM Comput. Surv., vol. 55, no. 2, Art. no. 42, pp. 1–96, Feb. 2022, doi: https://doi.org/10.1145/3485128.
[14] R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction. Cambridge, MA, USA: MIT Press, 1998.
[15] D. Silver et al., “A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play,” Science, vol. 362, no. 6419, pp. 1140–1144, Dec. 2018, doi: https:// doi.org/10.1126/science.aar6404.
[16] M. Hausknecht and P. Stone, “Deep recurrent Q-learning for partially observable MDPs,” in Proc. AAAI Fall Symp. Sequential Decision Making for Intelligent Agents, Arlington, VA, USA, Nov. 2015, pp. 29–37, Tech. Rep. FS-15-06.
[17] J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Proximal policy optimization algorithms,” arXiv, arXiv:1707.06347, July 2017, doi: https://doi.org/10.48550/arXiv. 1707.06347.
[18] V. Mnih et al., “Human-level control through deep reinforcement learning,” Nature, vol. 518, no. 7540, pp. 529–533, Feb. 2015, doi: https://doi.org/10.1038/nature14236.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 Sana Irshad, Muhammad Mateen Sadiq
This work is licensed under a Creative Commons Attribution 4.0 International License.
UMT-AIR follow an open-access publishing policy and full text of all published articles is available free, immediately upon publication of an issue. The journal’s contents are published and distributed under the terms of the Creative Commons Attribution 4.0 International (CC-BY 4.0) license. Thus, the work submitted to the journal implies that it is original, unpublished work of the authors (neither published previously nor accepted/under consideration for publication elsewhere). On acceptance of a manuscript for publication, a corresponding author on the behalf of all co-authors of the manuscript will sign and submit a completed the Copyright and Author Consent Form.
