Hazard maps are essential for public risk communication, preparedness, and evacuation planning, yet many local governments and communities lack professional cartographic expertise, resulting in inconsistent map quality. To address this gap we introduce Agentic Hazard Mapper, a multi‑agent system that automatically produces hazard maps by emulating professional cartographic workflows and adapting design choices to varying data, locations, and hazard contexts.
The system consists of three cooperating agents:
- Data‑processing agent ingests raw hazard data (e.g., flood depth, earthquake intensity), performs spatial interpolation, rasterization, and attribute normalization.
- Design‑decision agent mimics a cartographer’s iterative trade‑offs, evaluating candidate visualizations based on principles such as color hierarchy, symbol density, and information layering. It selects the optimal layout via a reinforcement‑learning reward function that balances spatial comprehension gains against information overload.
- Output agent renders the final layout into GIS‑compliant vector or raster files, automatically generating legends, scale bars, and annotations.
We first conducted a formative study with map experts to decompose the workflow, gather interaction requirements, and define evaluation metrics. A subsequent user study compared maps generated by the system with existing municipal maps across two tasks: spatial understanding and evacuation‑route planning. Results showed that system‑generated maps improved spatial accuracy by ~18%, reduced perceived workload (NASA‑TLX drop of 1.2 points), and shortened task completion time by an average of 23%. Participants consistently preferred the AI‑produced maps.
Follow‑up interviews highlighted the need for human oversight: experts want a review checkpoint to catch potential automation bias, and different user groups (professionals vs. the general public) require adjustable information density and symbol explanations. We discuss extending agentic support to the entire risk‑communication pipeline—from pre‑disaster outreach to real‑time alerts and post‑event assessment—and suggest integrating natural‑language generation to produce audience‑specific explanatory text.
Review: Agentic Hazard Mapper demonstrates the feasibility of multi‑agent automation in professional cartography, especially in enhancing map readability and reducing manual effort. However, further work is needed on transparency, interpretability, and cross‑cultural adaptation to unlock its full potential for risk communication.