Advances in coupling machine learning with hydrological simulation: A review

Abstract

Accurate and efficient hydrological simulation is critically important to sustainable water resources management amidst escalating climate change. As an indispensable scientific tool, hydrological modeling employs mathematical frameworks and computational techniques to quantitatively characterize hydrological processes, thereby playing a vital role in water resources assessment, the prediction and management of extreme hydrological events, and climate change impact evaluation. This review article systematically synthesizes recent advances in traditional hydrological models while critically examining their inherent methodological limitations. It further delineates the evolutionary trajectory of machine learning (ML) techniques in hydrological simulation and highlights the comparative advantages of data-driven ML approaches over conventional paradigms. Through a rigorous analysis of contemporary research, this review article establishes that coupling physically-based hydrological models with data-driven ML architectures represents the most promising pathway for overcoming fundamental bottlenecks in hydrological simulation. Furthermore, this review article concludes by identifying persistent challenges within existing coupling frameworks and projecting key future research directions in this rapidly evolving field.