GeoGPT for action ready flood and disaster risk geo-intelligence in Florida

Extreme weather in Florida can result compound impacts, crop damage, prolonged waterlogging, and inundation, that disrupt farm activities and complicate field scale assessment. Following an event, extension agents and growers typically need information on short timelines related to crop damage assessment to prioritize scouting, report impacts, and support recovery decisions, and flood-prone area information to anticipate where standing water and access constraints will persist and where follow-up interventions should be targeted. However, producing these products from Earth observation (EO) analysis-ready data (ARD) often requires fragmented geospatial tools, intensive preprocessing, and repeated iterations that delay action.

We present GAIA Bot, a conversational AI-based geospatial assistant piloted in Florida with extension agents and growers to convert EO ARD into action-ready information (ARI) for post-event decision support. In the Florida pilot workflow, users can interact with GAIA Bot through natural-language questions (e.g., “Which fields show likely damage since the storm?”; “Where are the flood-prone low areas that may remain saturated?”; “How does my field compare to the same period in prior year?”). GAIA Bot translates each request into an executable sequence that integrates publicly available spaceborne (e.g. Sentinel-2) observations with contextual geospatial layers (e.g., terrain and drainage proxies) and AI classifiers to generate field-scale damage indicators and priority scouting hotspots, flood-prone area maps that inform access and recovery planning, along with concise explanations for stakeholder communication.

Operational testing with end growers and extension agents indicates significant time savings relative to traditional multi tool approaches, enabling faster product generation and more frequent updates as new satellite observations become available. To support trustworthy decisions, we also explore a reasoning mechanism that produces structured evidence trails.