Bringing Agentic Search to Earth Observation Data Discovery

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

Summary

NASA has developed an agentic search system to simplify the discovery of geoscience datasets and tools within its vast Earth Observation archives. This public service system takes natural-language research queries and returns matching resources, addressing the challenge even domain experts face. The system leverages the NASA Earth Observation Knowledge Graph (NASA-EO-KG) to create NASA-EO-Bench, an open benchmark comprising 47,000 query-dataset pairs, including 21,000 task-based queries. A neural scorer, fine-tuned on NASA-EO-Bench, significantly outperforms cosine and BM25 baselines. Further integration with BM25 through score fusion enhances both Recall@10 and MRR by over 5x. Additionally, a zero-shot agentic reranking stage, requiring no extra training, elevates MRR by 28% on a stratified N=200 subset, highlighting the synergistic role of large language model reasoning with supervised retrieval.

Key takeaway

For Machine Learning Engineers tasked with enhancing complex data discovery, this work demonstrates a powerful hybrid approach. You should consider integrating knowledge graphs with supervised retrieval models, then applying zero-shot agentic reranking to boost performance. This strategy, shown to increase MRR by 28% and Recall@10 by over 5x, offers a clear path to significantly improve the precision and recall of your search systems for specialized datasets like Earth Observation data.

Key insights

Agentic search, amplified by knowledge graphs and LLM reranking, significantly improves Earth Observation data discovery.

Principles

• Knowledge graphs amplify value via agentic search.
• LLM reasoning complements supervised retrieval.
• Score fusion boosts retrieval metrics significantly.

Method

The system combines a neural scorer fine-tuned on NASA-EO-Bench with BM25 via score fusion, then applies zero-shot agentic reranking for enhanced retrieval.

In practice

• Use agentic search for complex data discovery.
• Develop benchmarks from existing KGs.
• Integrate LLM reranking for improved MRR.

Topics

• Agentic Search
• Earth Observation
• Knowledge Graphs
• Large Language Models
• Information Retrieval
• Data Discovery

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

Related on AIssential

• A Late Chunking Approach for Visual Documents, Does Agentic Search Make GraphRAG Obsolete? and More! — New research enhances LLM retrieval, recommendation systems, and RAG through novel architectures and training methods.
• OpenEarthAgent: A Unified Framework for Tool-Augmented Geospatial Agents — OpenEarthAgent enables geospatial AI agents to reason over satellite imagery using tool-augmented, multi-step logic.
• Multi-Sourced, Multi-Agent Evidence Retrieval for Fact-Checking — WKGFC unifies structured KGs and web evidence via an LLM agent for adaptive, self-improving fact-checking.
• SGR-Bench: Benchmarking Search Agents on State-Gated Retrieval — Search agents significantly underperform on state-gated retrieval due to failures in establishing correct site-specific states.
• ScholarQuest: A Taxonomy-Guided Benchmark for Agentic Academic Paper Search in Open Literature Environments — ScholarQuest benchmark reveals LLM-based search agents significantly underperform in academic literature exploration, despite outperforming baselines.
• META’s New SIRA: Superintelligence RAG — SIRA uses an LLM's parametric memory and BM25 for deterministic, single-shot information retrieval.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.