Problem
Research questions and friction points this paper is trying to address.
geo-intelligence
disaster reasoning
remote sensing
spatial reasoning
evidence-backed decision
GeoDisaster: Benchmarking Orchestrated Agents for Operational Disaster Geo-Intelligence
🤖 AI Summary
Existing remote sensing vision-language models struggle to support the tool invocation, spatial reasoning, and structured decision-making required for disaster geospatial intelligence. To address this gap, this work proposes GeoDisaster—the first benchmark tailored for operational-level disaster geospatial intelligence—encompassing 43 question types across five core tasks and integrating multisource remote sensing and GIS data. Furthermore, we introduce a multi-agent collaborative framework grounded in Role-Contract Expectation Alignment (RCEA), enhanced with failure-aware fine-tuning and contract-based reinforcement learning to improve evidential reasoning and state consistency. Experimental results demonstrate that GeoDisaster effectively exposes critical shortcomings of current models in tool utilization and evidence grounding, while the RCEA framework substantially enhances decision accuracy and reliability in disaster scenarios.
📝 Abstract
Remote-sensing vision-language models (RS-VLMs) have advanced Earth-observation analysis toward visual interpretation and instruction-following, yet fall short of operational geo-intelligence, which demands tool-grounded spatial reasoning and structured, evidence-backed decisions. We introduce GeoDisaster, an operational geospatial disaster reasoning benchmark with 2,921 verified instances across 43 question types and five task families: deforestation monitoring, multi-hazard analysis, building-damage assessment, flood-safe routing, and Sentinel-1 SAR flood monitoring. Instances integrate heterogeneous EO/GIS evidence-optical and SAR imagery, raster masks, vector geometries, road networks, and exposure layers-spanning hazard detection, damage assessment, exposure estimation, and diagnostic report generation. Ground-truth answers are grounded in executable geospatial workflows and deterministic consistency checks, removing the need for language-model annotation. We further propose an orchestrated multi-agent framework with 18 disaster-oriented tools, where role-specialized agents coordinate through explicit execution contracts, aligned via Role-Contract Expectation Alignment (RCEA): failure-aware supervised fine-tuning combined with contract-grounded reinforcement learning over dense step-level signals. Experiments show that GeoDisaster challenges existing RS-VLMs and agentic systems, while RCEA improves tool use, evidence grounding, state consistency, and decision generation.
Innovation
Methods, ideas, or system contributions that make the work stand out.
GeoDisaster
orchestrated multi-agent framework
Role-Contract Expectation Alignment (RCEA)
operational geo-intelligence
geospatial reasoning
🔎 Similar Papers
No similar papers found.
