Co-developing research-assisting AI for water resources professionals: the Digital Waters Flagship’s digital methods course

The accelerating complexity of global water challenges—driven by hydrologic intensification, a growing and urbanizing population, and proliferation of observational data—demands a new generation of water‑domain researchers who are both computationally fluent and capable of critically integrating artificial intelligence (AI) into scientific workflows. Yet, most geoscience doctoral programs provide limited training in open, reproducible computational methods, and generic AI tools often underperform in specialized environmental domains while lacking transparent attribution of sources. To address these gaps, the Digital Waters Flagship initiative designed and implemented an innovative doctoral‑level course that integrates open‑science software training with student‑driven co‑development of a domain‑adapted large‑language model (LLM) for hydrologic research assistance.

The course employs a flipped‑classroom model within the Digital Waters Virtual Research Environment (VRE), where students learn standardized, reproducible workflows using a repository structure composed of six core elements spanning data access, processing, modeling, visualization, and computational environments. Exceptional student repositories are publicly disseminated as open digital water use cases. In parallel, doctoral researchers participate in the co‑design of a hydrology‑focused research chatbot, DIWA ReChat, which is trained on authentic student‑generated workflow components and equipped with automatic knowledge‑source attribution to ensure transparency and proper crediting of contributions.

Course outcomes are evaluated through (1) pre‑/post‑assessment of computational competency, (2) evidence of improved reproducibility enabled by shared VRE infrastructure, and (3) empirical improvements in domain‑adapted LLM performance based on both conventional accuracy metrics and student‑designed AI efficacy criteria. Together, the course and chatbot development process demonstrate a scalable model for integrating open‑science education with responsible, domain‑aware AI tool creation. This work highlights a pathway for cultivating computationally capable researchers who can both leverage and critically evaluate AI in support of robust, transparent hydrologic science.