Error correction-based forecasting of reservoir water levels: Improving accuracy over multiple lead times

Abstract

Providing reliable reservoir water level forecasts is a challenge because of the accumulative errors in hydrological and reservoir routing models. We present a novel forecasting model that addresses these issues. The model consists of a hydrological model to simulate inflow, a reservoir routing model to simulate water levels, and an autoregressive model for error correction. The parameters for the hydrological model were calibrated with the objective of forecasting water levels over multiple lead times, while a back-fitting algorithm was used to recalibrate the parameters sequentially for the hydrological and autoregressive models. The results show that: (1) the forecasting performance of effective lead times can be enhanced by minimizing the difference between the forecasted and observed water levels for multiple lead times; (2) the most recent errors method is better than the one-step-ahead recursive prediction method; and (3) the back-fitting algorithm is superior to the joint inference method.