ClimatePolicyGen: a multi-agent framework for climate policy generation using multivariate and multimodal time series inputs

Effective climate policy formulation requires integrating multivariate, multi-view time-series data to develop actionable insights. However, existing approaches struggle with data synthesis, contextual alignment, and coherence in automated policy generation. This study introduces ClimatePolicyGen, a multi-agent framework leveraging large language models (LLMs) to automate climate policy development. The framework employs domain-specific agents to analyze environmental, socio-economic, and infrastructure trends, synthesizing structured policy recommendations. Experimental results demonstrate that ClimatePolicyGen surpasses baseline models, achieving a 12.3% improvement in coherence and an 18.7% increase in relevance, as validated by GEval and BERTScore. A case study on national climate strategies highlights its adaptability across diverse policy contexts. By enabling data-driven, adaptive, and region-specific policymaking, ClimatePolicyGen enhances global climate resilience and provides a foundation for data-driven policy drafting, with results validated through automated metrics as a first step toward practical deployment.