Seasonal Inflow Forecasts Using Gridded Precipitation and Soil Moisture Information: Implications for Reservoir Operation

Abstract

Reservoir inflow forecasts are important for guiding reservoir operation. This study proposes an integrated framework of incorporating different forms of seasonal inflow forecasts in identifying the optimal releases policy. Gridded precipitation forecasts from climate models have been widely used for forecasting inflow. Both precipitation forecasts and soil moisture estimates are used as predictors to provide one-season-ahead reservoir inflow forecasts by constructing a regression problem. Principal component analysis is used to reduce the dimension of the regression problem, and a Bayesian regression technique is employed to generate various forms of inflow forecasts such as deterministic, probabilistic and ensemble forecasts. Two optimization models are constructed to couple with different forms of inflow forecasts. The first model aims to maximize hydropower generation and the second one aims to minimize end-of-season reservoir storage deviation from the target storage. Both single-value inflow and ensemble forecasts are incorporated to find the optimal water releasing policy considering inflow uncertainty and end-of-season reservoir storage requirement. The proposed methodology is demonstrated for Huangcai Reservoir in southern China. Bayesian regression technique shows good performance of seasonal inflow forecasts with a Pearson correlation of 0.8 and rank probability score of 0.4, which outperforms climatology. The coupling of ensemble inflow forecasts and optimization models provides water managers a set of release policies considering inflow uncertainty.