Authors: Sujan Chandra Mondol, Daniel P. Ames – Brigham Young University
Presentation Type: Poster Presentation
Title: Empowering Decision-Making through LLM-Driven Autonomous Hydrologic Data Science
Presentation Type: Poster Presentation
Abstract: Although a variety of hydrologic data, from in situ to remote sensing to model-simulated data, are served through CIROH-affiliated data services, transforming these data into intelligible information remains a challenge that needs rigorous specialized expertise and time. This study presents workflows that leverage National Water Model (NWM) data and implement cloud-based guided data engineering and large-language-model-orchestrated, autonomous hydrologic data science to generate derived data products and intelligible information. The cloud-based data engineering pipeline computes streamflow indices from the NWM retrospective 3.0 dataset using big-data processing. The developed Model Context Protocol (MCP) provides the required interface for the large language model (LLM) to the NWM datasets. Supplementing the LLM units with a range of hydroinformatics tools, data, and instructions transformed them into specialized agents. By combining them within an agentic framework, the system can perform a set of analytical and interpretative workflows to generate ready-made information from the data. This study thus provides useful hydrologic information and insights for general guidance and decision-making, drawing on the latest operational data and historical patterns.