Characterizing the permeability variation in fractured rocks is important in various subsurface applica-tions, but how the permeability evolves in the foundation rocks of high dams during operation remains poorly understood. This permeability change is commonly evidenced by a continuous decrease in the amount of discharge (especially for dams on sediment-laden rivers), and can be attributed to fracture clogging and/or hydromechanical coupling. In this study, the permeability evolution of fractured rocks at a high arch dam foundation during operation was evaluated by inverse modeling based on the field time-series data of both pore pressure and discharge. A procedure combining orthogonal design, transient flow modeling, artificial neural network, and genetic algorithm was adopted to efficiently estimate the hydraulic conductivity values in each annual cycle after initial reservoir filling. The inverse results show that the permeability of the dam foundation rocks follows an exponential decay annually during oper-ation (i.e. K/K0 = 0.97e-0.59t + 0.03), with good agreement between field observations and numerical simulations. The significance of the obtained permeability decay function was manifested by an assessment of the long-term seepage control performance and groundwater flow behaviors at the dam site. The proposed formula is also of merit for characterizing the permeability change in riverbed rocks induced by sediment transport and deposition.