When
Where
SPECIAL SEMINAR - DIFFERENT DAY, TIME, & LOCATION - Available in person and via Zoom (see email for link)
Abstract
Artificial intelligence is increasingly used across the geosciences, but most applications still treat AI primarily as a tool for prediction, classification, or data analysis. Recent advances in large language models and AI agents are beginning to change this role. By combining reasoning with access to scientific models, data, and computational tools, agentic AI systems can potentially participate in multiple stages of a scientific workflow—planning analyses, operating models, evaluating results, and deciding what to do next.
Hydrologic modeling provides a useful setting for exploring these emerging capabilities. Recent work demonstrates how AI agents can connect observations, process-based models, and computational environments to automate complex modeling workflows with limited human intervention. Beyond automation, large language models can also be used to diagnose model behavior and guide model calibration, using hydrologic knowledge and model responses to inform sequential decisions rather than relying solely on conventional numerical optimization.
These developments point toward a broader role for AI in geoscience: moving from tools that generate predictions toward systems that can assist with scientific reasoning and investigation. Important challenges remain, including physical consistency, reliability, reproducibility, and the effective incorporation of domain knowledge. Addressing these challenges will be essential for determining how agentic AI can complement human expertise and become a useful component of future geoscientific research.
Bio
Songkun Yan is a Ph.D. student in Environmental Science in the School of Civil Engineering and Environmental Science at the University of Oklahoma. His research focuses on agentic AI for hydrology, with particular interests in integrating large language models, scientific reasoning, and process-based hydrologic models to support modeling and scientific investigation. His recent work explores AI agents for automated hydrologic modeling and LLM-based diagnostic reasoning for model calibration and parameter inference. Before joining the University of Oklahoma, he received both his B.S. in Geographic Information Science and M.S. in Photogrammetry and Remote Sensing from Peking University.
For questions, please contact Hoshin Gupta (hoshin@arizona.edu).