Earth observation question answering improved with observation-support graph reasoning
OSGR: Observation-Support Graph Reasoning for Earth Observation Question Answering
Artificial IntelligenceInformation Retrieval
Summary
Answering questions about Earth from satellite images is hard because it needs searching many places and times carefully. The authors propose a new method called OSGR that breaks the problem into two steps: first figuring out concepts and their relationships, then matching those to actual data. This makes the search smarter and faster. Their method works better than existing approaches on test problems, improving accuracy and using fewer steps.
What this means in practice
- •For earth observation analysts: Enable more accurate and efficient querying of satellite data archives for geographic insights under complex spatiotemporal limits.
- •For disaster response teams: Support faster identification and reasoning about relevant satellite observations to assist disaster management decisions.
- •For geospatial intelligence developers: Develop advanced tools that integrate symbolic reasoning with satellite imagery to improve automated geographic question answering services.$Commercial implications: This paper enables new geospatial AI products that offer improved satellite data query capabilities for government and commercial clients.
Authors
Dazhou Yu, Zerong Wang, Riyang Bao, Longxiang Li, Yang Liu, Liang Zhao
Abstract
Answering geographic queries over Earth observation (EO) archives requires solving a joint retrieval-and-derivation problem under strict spatiotemporal constraints. Existing LLM agents fail to ground observation items or suffer from search-space explosion. We propose OSGR, a two-stage framework that factorizes this search into offline symbolic concept deduction and online Relation-Conditioned Grounding. OSGR leverages an offline Observation-Support Graph to deduce valid target-to-result support paths at the concept level, and applies spatiotemporal compatibility kernels to instantiate grounded observation-item sets. On Earth-Bench and DisasterM3-Reason, OSGR improves GPT-4o accuracy from 49.3\% to 63.7\% while reducing the execution-step ratio from 2.9 to 2.5.