GeoAI Platform for Natural Language Exploration of Multidimensional Flood Risk Scenarios

Abstract While many coastal jurisdictions increasingly use flood models that incorporate future climate change projections, floodplain managers, planners and community organizers frequently need to answer questions that draw on multiple data sources and model outputs. A single compound planning question requires stitching together results from inundation outputs, infrastructure records, and so- cioeconomic datasets that currently live in disconnected silos. The need for complex workflows and specialized tooling constrains the questions practitioners can ask and the speed with which they can get the results. This paper presents a geospatial AI platform that addresses this gap through three architectural contributions: (i) A multidimensional scenario store that organizes 141 flood scenarios within a single queryable structure along with other physical and socioeconomic datasets; (ii) A deterministic tool-routed LLM agent that maps compound natural language queries into geospatial tool calls; and (iii) A bidirectional map-chat interface that preserves spatial grounding throughout the analytical workflow.