AI-Powered Climate Policy

The literature review highlights the escalating environmental degradation and its profound impact on human well-being. Climate change, pollution, and biodiversity loss are primary concerns. The section emphasizes the critical link between environmental policy and human well-being, citing studies on health impacts from air pollution and extreme weather events.

The Delphi method is presented as a valuable tool for foresight studies, enabling experts to build consensus on complex future issues and develop scenarios. The emergence of GPT models in decision-making is also discussed, noting their ability to generate human-like text and assist in tasks like content creation and analysis of text-based data.

The proposed methodology integrates GPT support within the Real-Time Spatial Delphi approach. It leverages a previous study on Dublin 2050 climate scenarios, optimizing its outcomes to generate policy drafts using ChatGPT-4. Key inputs for policy extraction include: Key factors (flood risk, coastal erosion, extreme weather events), Geographic coordinates from expert inputs, Convergence circles, and Experts' comments.

Two main prompts were used: one for environmental policies and one for people's well-being. A statistical coefficient Ipol,j = α ⋅ Ds,j + β ⋅ Ic,j, combining spatial density (Ds) and comment importance (Ic), is used to score policies. A threshold of Ipol,j < 0.7 discards less relevant policies, ensuring strong spatial and textual relevance for the final 20 policies per prompt.

The methodology successfully extracted 20 policies for each of the three climate scenarios (flood risk, coastal erosion, extreme weather events) and two prompts (ecosystem-based, citizen-oriented). The policies emphasize infrastructural resilience, nature-based solutions, and socio-economic preparedness.

• Flood Risk Management: High priority on multi-layered flood barriers, upgraded drainage systems, urban wetlands, and flood insurance expansion.
• Coastal Erosion Management: Focus on offshore breakwaters, artificial reefs, beach nourishment, and coastal relocation planning.
• Extreme Weather Events: Strategies like urban green spaces, tree canopies, early warning systems, and financial aid for victims.

Descriptive statistics and Pearson correlation analysis (up to 0.978 for S3 P2) validated the robustness and coherence of the extracted policies, indicating strong alignment between spatial relevance and expert textual justification.

This study successfully integrates AI-assisted policy generation into Delphi-based scenario development, bridging the gap between foresight and actionable recommendations. The framework offers a scalable, semi-automated, and expert-informed approach for climate adaptation planning.

Methodologically, it enhances Delphi by introducing real-time spatial integration, AI-driven policy drafting, and quantitative validation through statistical coefficients and correlation metrics. Practically, it reduces cognitive and resource burdens, making large-scale participatory scenario processes more accessible.

Limitations include GPT's potential lack of true contextual awareness and the linear compensatory aggregation of scores. Future research will explore pilot projects, hybrid AI-human systems, multi-criteria analysis, and model configuration variations to further refine the approach.