Integrating Kalman smoother and hydrodynamic model for accurate estimation of true inflow for river-type reservoirs

This is the authors' abstract. We don't add key points for this paper.

Study region The Three Gorges Reservoir, China. Study focus Reservoir theoretical “true” inflow estimation is crucial for forecasting and operational decisions. However, in river-type reservoirs where the dam is far from the upstream boundary, flow propagation time lag and information loss reduce inflow estimation accuracy. The study proposed a novel method integrating a Kalman smoother with a 1D hydrodynamic model to estimate theoretical “true” inflow under time-lag conditions. The state transition equations for inflow and water level were developed using a Gaussian random walk and a one-dimensional hydrodynamic model. The hydrodynamic model was executed under varying inflow ensemble members to simulate water levels and flow signal propagation time. The Kalman smoother was employed to update generated inflow ensemble. The proposed method was validated by comparing with water balance (WB), analytical method (AM), and ensemble Kalman filter (EnKF). New hydrological insights for the region (1) The proposed method improved peak flow estimation accuracy by 2∼3%, reduced RMSE by 40∼70% under synthetic experiments, and increased the lateral inflow NSE from 0.717 to 0.909; (2) the effect of time lags was reduced through dynamic estimation of the flow propagation time and backward propagation of information in both synthetic and real cases, while WB and EnKF underperformed. The proposed method provides a reliable solution for theoretical “true” inflow estimation, supporting more effective reservoir management.