EO-Agents: A Three-Agent LLM Pipeline for Earth Observation Hypothesis Generation

2026-07-02 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Environmental Science & Earth Systems · Depth: Advanced, quick

Summary

EO-Agents introduces a novel three-agent large language model (LLM) pipeline designed for generating scientific hypotheses in Earth observation. Unlike prior work relying on unstructured literature, this system directly grounds its hypothesis generation in the NASA Earth Observation Knowledge Graph. The pipeline integrates a heterogeneous graph neural network, trained on historical co-usage relations, to rank candidate dataset pairings. Subsequently, a three-agent LLM system filters, generates, and evaluates structured research hypotheses. When applied to 1,475 NASA datasets, EO-Agents successfully produced 160 hypotheses across diverse Earth-science domains, including ecohydrology and glaciology. The system's predicted novel dataset pairings were rated as nearly plausible as actual co-usages from literature, demonstrating its ability to surface scientifically coherent yet unexplored combinations. Furthermore, a 2*2*2 factorial experiment using GPT-5.2 and Claude Sonnet 4.6 revealed stable hypothesis rankings but strong dependence of absolute scores on judge identity, highlighting LLM evaluation challenges.

Key takeaway

For research scientists exploring novel scientific hypotheses, EO-Agents demonstrates a robust method for utilizing structured knowledge graphs and multi-agent LLMs. You should consider integrating graph neural networks with LLM pipelines to ground hypothesis generation in factual data, moving beyond unstructured text. Be aware that LLM evaluation scores can be highly judge-dependent; therefore, employ diverse evaluation strategies or multiple judges to ensure reliable assessment of generated scientific claims. This approach can surface coherent, unexplored dataset combinations.

Key insights

EO-Agents uses a GNN and a three-agent LLM pipeline to generate scientifically plausible Earth observation hypotheses from structured knowledge graphs.

Principles

• Grounding LLM hypothesis generation in structured KGs improves relevance.
• Heterogeneous GNNs can effectively rank scientific dataset co-usage.
• LLM evaluation scores can vary significantly by judge identity.

Method

The pipeline involves a heterogeneous graph neural network for ranking candidate dataset pairings, followed by a three-agent LLM system that filters, generates, and evaluates structured research hypotheses.

In practice

• Apply GNNs to knowledge graphs for scientific relation discovery.
• Use multi-agent LLM systems for structured hypothesis generation.
• Consider judge variability when evaluating LLM-generated scientific content.

Topics

• Earth Observation
• Hypothesis Generation
• Large Language Models
• Graph Neural Networks
• Knowledge Graphs
• LLM Evaluation

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Risk-Aware LLM Agents for Geospatial Data Retrieval: Design and Preliminary Adversarial Evaluation — The framework uses LLM agents to convert natural language into structured API calls for geospatial data retrieval, integrating safety guardrails.
• AI scientists produce results without reasoning scientifically — LLM-based scientific agents execute workflows but lack fundamental scientific reasoning patterns, primarily due to the base model.
• Integrating Graphs, Large Language Models, and Agents: Reasoning and Retrieval — Integrating graphs with LLMs enhances reasoning, retrieval, and decision-making by combining structured knowledge with semantic understanding.
• [AINews] OpenAI GPT-next disproves 80 year old Erdős planar unit distance problem for under $1000 — General-purpose LLMs are achieving significant scientific breakthroughs and driving architectural innovation.
• TerraBench: Can Agents Reason Over Heterogeneous Earth-System Data? — TerraBench integrates LLM reasoning with scientific tools to unify complex Earth-system data analysis, enabling agents to reason over heterogeneous environmental inputs.
• AI Trends 2026: OpenClaw Agents, Reasoning LLMs, and More [Sebastian Raschka] - 762 — LLM evolution focuses on post-training reasoning, tool use, and inference scaling to enhance problem-solving and efficiency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.