Evolution and prospects of artificial intelligence-driven research paradigms in watershed hydrology

Under the combined influence of climate change and human activities, the complexity of watershed water systems has intensified, and traditional hydrological research methods face new challenges. Artificial intelligence (AI) technologies, represented by big data and deep learning, have developed rapidly and profoundly transformed the research and application paradigms of watershed hydrology. This study reviews the evolution of watershed hydrological research from empirical and statistical methods to mechanistic models and data-driven paradigms. It focuses on the research progress of AI in fields including complex hydrological process simulation, multi-source observation and fusion, and digital twin watershed construction. Overall, the advantages of AI in high-dimensional feature extraction, nonlinear relationship identification, and cross-watershed knowledge transfer significantly enhance the accuracy, generalization, and rapid response capability of hydrological simulation, providing new perspectives and tools for understanding complex hydrological systems and coping with environmental change. However, the in-depth application of AI in watershed hydrology faces several challenges, including the inherent sparsity and heterogeneity of hydrometeorological data, insufficient physical consistency and extrapolation robustness of models, and operational integration. In the future, promoting the deep integration of data-driven methods and hydrological mechanisms, developing a watershed system research framework supported by multi-source information, and building an interdisciplinary research and application ecosystem are key directions for AI to empower innovation in watershed hydrology and support the construction of national water networks and the development of smart water conservancy.