Bringing Agentic Search to Earth Observation Data Discovery
2026-07-02 • Information Retrieval
Information RetrievalMachine Learning
AI summaryⓘ
The authors created a smart search tool for NASA's geoscience data that understands natural language questions and finds the best matching datasets and tools. They used a large knowledge graph and a specially trained computer model to improve search accuracy much more than traditional methods. They also added a stage that uses large language models to further improve results without extra training. This shows combining knowledge graphs with advanced AI can help experts find data more easily.
NASA Earth ObservationKnowledge GraphNatural Language ProcessingLarge Language ModelsInformation RetrievalNeural ScorerBenchmark DatasetRecall@10Mean Reciprocal RankBM25
Authors
Minghan Yu, Youran Sun, Chugang Yi, Yixin Wen, Haizhao Yang
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.