Estimation of reservoir inflow considering water level observation errors and dynamic reservoir capacity

Accurate reservoir inflow estimation is the foundation of flood forecasting and control. However, traditional inflow estimation methods based on water balance ignore the effects of observational errors in water level and dynamic reservoir capacity. These methods will cause significant fluctuation, generating negative inflow values over short intervals, and cannot capture lateral flows. To address these challenges, the study proposes a novel reservoir inflow estimation method combining the Rauch-Tung-Striebel (RTS) smoother to account for water level observation errors and a 1D hydraulic model (1D-HM) to account for dynamic reservoir capacity. Firstly, a potential inflow ensemble is stochastically generated. Then, multiple 1D-HMs are conducted to simulate water levels under different potential inflow scenarios. Finally, the RTS smoother is employed to update the inflow ensemble and historical inflow records based on the differences between observed and simulated water levels. The estimation of smoothed upstream inflow and lateral inflow is achieved through rolling filtering. The proposed method is validated through numerical experiments and a real-world case study of the Three Gorges Reservoir. The results show that: (1) In numerical experiments, the proposed method outperforms other comparative methods under various conditions, including errors in water levels, dynamic reservoir capacities, and lateral flows. (2) In the real case study, the proposed method can generate no-fluctuation reservoir inflow and lateral flow estimates at 15-minute intervals.