Bringing Agentic Search to Earth Observation Data Discovery

2026-07-02

AI Summary

Earth science researchers face significant challenges in efficiently retrieving relevant data and tools from NASA’s vast Earth observation archives. To address this, this work proposes the first intelligent agent-based search system that integrates a knowledge graph with a large language model to enable precise resource discovery through natural language queries. The study introduces two key contributions: the creation of NASA-EO-Bench, an open benchmark comprising 47,000 query–data pairs, and a novel zero-shot agent-based reranking mechanism. By combining BM25 retrieval, neural scoring models, and a score fusion strategy, the system achieves more than a fivefold improvement over baseline methods in both Recall@10 and Mean Reciprocal Rank (MRR). Furthermore, the proposed zero-shot reranking approach yields an additional 28% gain in MRR, demonstrating its effectiveness in enhancing retrieval accuracy without task-specific training.

Abstract

NASA and its data centers hold thousands of geoscience datasets and tools like Worldview, Giovanni, the Science Discovery Engine, and Harmony. Finding the right one is hard even for domain experts. We present an agentic search system, deployed as a public service for the geoscience community, that takes a natural-language research query and returns the matching datasets and tools. We demonstrate that, in the era of large language models, the latent value of knowledge graphs (KGs) can be substantially amplified through agentic search. From the NASA Earth Observation Knowledge Graph (NASA EO-KG) we derive NASA-EO-Bench, an open benchmark of 47k query-dataset pairs (21k task-based queries). A neural scorer fine-tuned on NASA-EO-Bench beats cosine and BM25 baselines. Further combining it with BM25 via score fusion raises both Recall@10 (R@10) and MRR by over 5x. On top of this supervised pipeline, we add a zero-shot agentic reranking stage that, without any additional training, lifts MRR by 28% on a stratified N=200 subset, showing that LLM reasoning is complementary to supervised retrieval.

Problem

• Earth Observation
• Data Discovery
• Agentic Search
• Knowledge Graph
• Natural Language Query

Innovation

• agentic search
• knowledge graph
• large language models
• information retrieval
• Earth observation