Toward Automated Scientific Discovery in Hydrology: The Opportunities and Dangers of AI Augmented Research Frameworks

Abstract

This commentary explores the potential of artificial intelligence (AI) to transform hydrological modelling workflows. We introduce a prototype AI-assisted framework called INDRA (Intelligent Network for Dynamic River Analysis) that leverages a multi-agent architecture composed of specialised large language models (LLMs) to assist in model conceptualization, configuration, execution, and interpretation. INDRA integrates with CONFLUENCE, a comprehensive modelling framework, to provide context-aware guidance and automation throughout the modelling process. We discuss the opportunities and dangers of AI-augmented research frameworks, emphasising the importance of maintaining human oversight while harnessing AI's potential to enhance efficiency, reproducibility, and scientific understanding. We argue that AI-assisted workflows could democratise advanced hydrological modelling, enabling researchers worldwide to address critical water resources challenges, particularly in understudied regions. While acknowledging potential biases and risks, we advocate for responsible AI integration to catalyse a new paradigm in hydrological science.