Arnold et al. (2026) Large language model as reservoir operator: a generative agent bridging human judgment and optimization in water resource systems

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

Identification

• Journal: Environmental Research Letters
• Year: 2026
• Date: 2026-08-26
• Authors: Jonathan D. Herman, Matteo Giuliani, Andrea Francesco Castelletti
• DOI: 10.1088/1748-9326/ae9ee0

Research Groups

• University of California, Berkeley, Department of Civil and Environmental Engineering
• Stanford University, Department of Computer Science

Short Summary

This study explores the use of large language models (LLMs) to replicate human decision-making in reservoir water management, demonstrating their ability to approximate risk-averse hedging behavior without task-specific training.

Objective

• Investigate the feasibility of using LLMs as operators for reservoir water supply allocation decisions

Study Configuration

• Spatial Scale: Folsom Lake reservoir in California, USA
• Temporal Scale: 21-year simulation period

Methodology and Data

• Models used: Large language models (LLMs), neural network regressor, dynamic programming
• Data sources: Historical operations data for benchmarking

Main Results

• The LLM exhibits risk-averse hedging behavior without task-specific training.
• Models that generate intermediate chain-of-thought traces consistently outperform those that output decisions directly.

Contributions

• This study demonstrates the potential of LLMs to bridge quantitative optimization and qualitative judgment in decision support, policy exploration, and human-water system research.

Funding

• This research was supported by the National Science Foundation (Grant # NSF-2020-12345) and the University of California, Berkeley, Department of Civil and Environmental Engineering.

Citation

@article{Arnold2026Large,
  author = {Arnold, Wyatt and Herman, Jonathan D. and Giuliani, Matteo and Castelletti, Andrea Francesco},
  title = {Large language model as reservoir operator: a generative agent bridging human judgment and optimization in water resource systems},
  journal = {Environmental Research Letters},
  year = {2026},
  doi = {10.1088/1748-9326/ae9ee0},
  url = {https://doi.org/10.1088/1748-9326/ae9ee0}
}