Probability Risk Assessment of Reservoir Operations

Reservoirs are operated and managed based on a set of rule curves. Given an ob- served time series of reservoir inflows, optimal reservoir operations can be determined by linear programming or other deterministic techniques of operation research. However, reservoir inflows are inherently stochastic and can be characterized by a random process. Thus, given a set of rule curves and a deterministic optimization technique, reservoir performance can still vary from one year to the next. In this study, we propose a stochastic approach for probabilistic risk assessment of reservoir operations for the Shihmen Reservoir in northern Taiwan. The approach consists of a stochastic reservoir inflow simulation model and a deterministic optimal reservoir operation technique, the network flow model. The stochastic reservoir inflow simulation model considers daily flows induced by ty- phoons and non-typhoon daily flows separately. Occurrences of typhoons are modeled by a Poisson process, and durations and event-total flows of typhoons are characterized by a bivariate gamma distribution. Log-transformed non-typhoon daily flows are modeled by an ARMA(1,1) time series process. Ten thousand realizations of annual daily flow series were simulated. The deterministic optimal reservoir operation was then applied to each of the 10,000 simulated annual series. Finally, a probabilistic risk assessment of reservoir operations was conducted using results of the 10,000 sets of optimal operations. It was found that the risk of agricultural water shortage was highest in February and lowest in September. The simulation results can also facilitate a drought frequency analysis