Multi-step ahead daily inflow forecasting using ERA-Interim reanalysis dataset based on gradient boosting regression trees

Abstract

Abstract. Inflow forecasting plays an essential role in reservoir management and operation. The impacts of climate change and human activities make accurate inflow prediction increasingly difficult, especially for longer lead times. In this study, a new hybrid inflow forecast framework with ERA-Interim reanalysis data as input, adopting gradient boosting regression trees (GBRT) and the maximum information coefficient (MIC) was developed for multi-step ahead daily inflow forecasting. Firstly, the ERA-Interim reanalysis dataset provides enough information for the framework to discover inflow for longer lead times. Secondly, MIC can identify effective feature subset from massive features that significantly affects inflow so that the framework can avoid over-fitting, distinguish key attributes with unimportant ones and provide a concise understanding of inflow. Lastly, the GBRT is a prediction model in the form of an ensemble of decision trees and has a strong ability to capture nonlinear relationships between input and output in long lead times more fully. The Xiaowan hydropower station located in Yunnan Province, China is selected as the study area. Four evaluation criteria, the mean absolute error (MAE), the root mean square error (RMSE), the Nash-Sutcliffe efficiency coefficient (NSE) and the Pearson correlation coefficient (CORR), were used to evaluate the established models using historical daily inflow data (1/1/2017–31/12/2018). Performance of the presented framework was compared to that of artificial neural networks (ANN), support vector regression (SVR) and multiple linear regression (MLR) models. The experimental results indicate that the developed method generally performs better than other models and significantly improves the accuracy of inflow forecasting at lead times of 5–10 days. The reanalysis data also enhances the accuracy of inflow forecasting except for forecasts that are one-day ahead.