A Generative Artificial Intelligence Framework for Earth Observation Analysis

Generative artificial intelligence (AI), specifically Large Language Models (LLMS) such as generative pretrained transformers (GPTS) have emerged as a transformational technology in modern research and applications. This technology will revolutionize a wide variety of fields, including remote sensing and Earth observation. We explore the use of LLMS to improve and automate EO analysis, making these results more accessible to a broad range of decision makers. We develop a framework using this technology to support the Committee on Earth Observation Satellites (CEOS) and NASA’s Open Science initiative. This framework leverages current work on Earth observation tools, including the Open Data Cube (ODC) [1] and the OEA Algorithm Hub (AlgoHub). [2]In our work, we investigate the ability of large language models to orchestrate and reason with tools created to access geospatial data using the Open Data Cube and the OEA Algorithm Hub [2]. In this paper we propose a natural language analytical framework for orchestrating Earth observation (EO) analysis. The framework uses large language models to interpret an analysis request, devise a solution based on ODC and AlgoHub tools, and execute the analysis. Our experimental investigations demonstrate that an LLM, with its embedded contextual knowledge, can effectively discern and sequence the necessary processing steps required to fulfill Earth observation analysis requests. This research demonstrates the potential of Generative AI (LLMs) in streamlining and enhancing EO analyses and empowering the observation community.In summary, Generative AI (LLMs) can enhance the capabilities of Earth observation analysis by automating tasks, providing natural language interfaces, and facilitating the extraction of valuable insights from the vast and complex datasets generated by remote sensing and EO technologies.