NASA launches agentic search for Earth Observation data discovery

NASA has deployed a public agentic search system designed to help the geoscience community find relevant datasets and tools through natural-language queries. This service leverages the NASA Earth Observation Knowledge Graph (NASA EO-KG) to address the difficulty of locating resources among thousands of geoscience assets.

• The system introduces NASA-EO-Bench, an open benchmark containing 47k query-dataset pairs, including 21k task-based queries.
• A neural scorer fine-tuned on this benchmark outperforms cosine and BM25 baselines.
• Combining the neural scorer with BM25 via score fusion increases Recall@10 (R@10) and MRR by over 5x.
• Adding a zero-shot agentic reranking stage improves MRR by 28% on a stratified subset of 200 queries without additional training.

The authors demonstrate that latent knowledge graph value is substantially amplified through agentic search, showing that LLM reasoning complements supervised retrieval methods.