[Research Article] An LLM-based multi-agent system for remote sensing analysis

A new study published in Big Earth Data proposes a large language model (LLM)-based multi-agent framework for remote sensing analysis that integrates data resources, analytical tools, and AI-driven reasoning to assist non-expert users in conducting complex remote sensing tasks. Using the ExpertsRS prototype built on AutoGen and DeepSeek-V3, experiments demonstrate that the system can transform ambiguous queries into expert-level analytical workflows and outperform baseline LLM and single-agent approaches in planning efficiency and result accuracy. The Sentinel-2 dataset used in this study is openly available from the European Space Agency at https://www.esa.int/Applications/Observing_the_Earth/Copernicus/Sentinel-2. The code repository for this work is accessible at https://github.com/zhelunSun/URSA.

Citation

Sun, Z., Zhou, Y., & Yang, J. (2026). An LLM-based multi-agent system for remote sensing analysis. Big Earth Data, 1–25. https://doi.org/10.1080/20964471.2025.2600178

Abstract

Remote sensing analysis converts remotely sensed data into actionable insights across diverse fields. However, traditional data-driven approaches frequently overlook the technical challenges encountered by end-users lacking remote sensing expertise. To address this issue, we proposed a multi-agent system framework based on a large language model, designed to facilitate remote sensing analysis for non-experts. The framework consists of three main modules: Data, which includes remote sensing datasets and supplementary information; Tools, encompassing algorithms and visualization tools; and Brain, which provides AI-driven task management and reasoning. We employed a prototype system called ExpertsRS, developed on the AutoGen framework and DeepSeek-V3, to validate the proposed framework. This prototype features three LLM-powered agents with distinct roles and functionalities, collaborating to fulfill user requests. The system was evaluated through two experiments. The first experiment demonstrated that ExpertsRS could translate ambiguous queries into structured analytical workflows and produce outputs comparable to those of human experts. The second experiment indicated that ExpertsRS outperformed both a baseline LLM and a single-agent configuration in planning efficiency and result accuracy, with only a modest increase in token usage. Our system underscores the potential of large language model-based multi-agent system to assist end-users in overcoming technical barriers in remote sensing analysis.

#geoscience #remote sensing #earth observation #GIS #data analysis #Big Data #visualization 

Big Earth Data is an interdisciplinary Open Access journal which aims to provide an efficient and high-quality platform for promoting the sharing, processing and analyses of Earth-related big data, thereby revolutionizing the cognition of the Earth’s systems. The journal publishes a wide range of content, including Research Articles, Review Articles, Data Notes, Technical Notes, and Perspectives. It is now included in ESCI (IF=3.8, Q1), Scopus (CiteScore=9.0, Q1), Ei Compendex, GEOBASE, and Inspec. Starting from 2023, Big Earth Data has announced a new award series for authors: Best and Outstanding Paper Awards.