Exploring the opportunities and challenges of using large language models to represent institutional agency in land system modelling

Zeng, Yongchao; Brown, Calum; Raymond, Joanna; Byari, Mohamed; Hotz, Ronja; Rounsevell, Mark

doi:https://doi.org/10.5194/esd-16-423-2025

Articles | Volume 16, issue 2

https://doi.org/10.5194/esd-16-423-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/esd-16-423-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 16, issue 2

Research article

|

13 Mar 2025

Research article |

| 13 Mar 2025

Exploring the opportunities and challenges of using large language models to represent institutional agency in land system modelling

Yongchao Zeng, Calum Brown, Joanna Raymond, Mohamed Byari, Ronja Hotz, and Mark Rounsevell

Download

Final revised paper (published on 13 Mar 2025)
Supplement to the final revised paper
Preprint (discussion started on 12 Mar 2024)

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2024-449', Anonymous Referee #1, 11 Apr 2024

In this manuscript, the authors describe their work inducing a large language model to “role-play” as various kinds of policy decisionmakers in an agent-based land use model. While a human operator needs to stay in the loop to keep the LLM on task and producing output in the correct format, the agents—when properly prompted—are capable of producing policy actions that achieve their goal. As befits such a novel method, the authors do more than just using the policy actions output by the model; they also dig in to the apparent “reasoning” behind its actions.
This is a fascinating piece of research. The paper is composed logically, well-written, and the figures are clear. However, I do have a number of comments, the most important of which relate to the manuscript’s eliding of how LLMs actually work. Once these are addressed, though, it will stand as an important, foundational contribution to the use of LLMs in agent-based land use modeling.
Please see the attached PDF for my comments.

Citation: https://doi.org/10.5194/egusphere-2024-449-RC1
- AC1: 'Reply on RC1', Yongchao Zeng, 24 Dec 2024
  
  Dear Reviewer,
  We are grateful for your prompt and thorough review of our manuscript. Your insightful feedback has been invaluable in refining our work.
  We also thank you for your patience. We initially planned to respond once all reviews were secured to address feedback comprehensively. However, as this research spans multiple disciplines, finding reviewers in this new domain was challenging. We are grateful to the journal’s editor for their efforts in securing reviews, enabling us to proceed.
  We are committed to addressing the comments thoroughly to ensure the work makes a meaningful contribution to this evolving field. In the PDF attached, we outline our approach to addressing your comments point by point.
  Sincerely,
  The authors
  
  Citation: https://doi.org/10.5194/egusphere-2024-449-AC1
CC1:
'Generalisability & scalability', Oliver Perkins, 13 Apr 2024

Dear authors,
I greatly enjoyed reading this impressive work. Please find enclosed some questions and comments, which relate primarily to the scalability and generalisability of what you have achieved here.
All best
Ol Perkins

Citation: https://doi.org/10.5194/egusphere-2024-449-CC1
- AC3: 'Reply on CC1', Yongchao Zeng, 24 Dec 2024
  
  Thank you for your comments. As per your suggestion, we have incorporated the detailed answers to your questions in "Reply on RC2".
  
  Citation: https://doi.org/10.5194/egusphere-2024-449-AC3
RC2:
'Comment on egusphere-2024-449', Oliver Perkins, 12 Nov 2024

Please see attached file. Please disregard my prior community comment, which has been incorporated into this new file.

Citation: https://doi.org/10.5194/egusphere-2024-449-RC2
- AC2: 'Reply on RC2', Yongchao Zeng, 24 Dec 2024
  
  Dear Reviewer,
  Thank you for your detailed review of our manuscript. Your feedback has provided valuable insights that will help us improve the clarity and impact of our work.
  As our research spans multiple disciplines, your expertise and constructive suggestions are especially appreciated. We have carefully reviewed your comments and are committed to addressing them thoroughly. In the PDF attached, we outline our responses to your major and minor comments.
  Sincerely,
  
  The authors
  
  Citation: https://doi.org/10.5194/egusphere-2024-449-AC2

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

ED: Publish subject to minor revisions (review by editor) (26 Dec 2024) by Ben Kravitz

AR by Yongchao Zeng on behalf of the Authors (11 Jan 2025) Author's response Author's tracked changes Manuscript

ED: Publish as is (14 Jan 2025) by Ben Kravitz

AR by Yongchao Zeng on behalf of the Authors (16 Jan 2025) Author's response Manuscript

Download

Article (2067 KB)
Full-text XML

Short summary

This study explores using large language models (LLMs) to simulate policy-making in land systems. We integrated LLMs into a land use model and simulated LLM-powered institutional agents steering meat production by taxation. The results show LLMs can generate boundedly rational policy-making behaviours that can hardly be modelled using conventional methods; LLMs can offer the reasoning behind policy actions. We also discussed LLMs’ potential and challenges in large-scale simulations.