Data-driven models for monthly streamflow time series prediction

C. L. Wu and K. W. Chau* 2 Dept. of Civil and Structural Engineering, Hong Kong Polytechnic University, 3 Hung Hom, Kowloon, Hong Kong, People’s Republic of China 4 5 *Email: [email protected] 6 ABSTRACT 7 Data-driven techniques such as Auto-Regressive Moving Average (ARMA), K-Nearest-Neighbors (KNN), and 8 Artificial Neural Networks (ANN), are widely applied to hydrologic time series prediction. This paper investigates 9 different data-driven models to determine the optimal approach of predicting monthly streamflow time series. Four 10 sets of data from different locations of People’s Republic of China (Xiangjiaba, Cuntan, Manwan, and 11 Danjiangkou) are applied for the investigation process. Correlation integral and False Nearest Neighbors (FNN) 12 are first employed for Phase Space Reconstruction (PSR). Four models, ARMA, ANN, KNN, and Phase Space 13 Reconstruction-based Artificial Neural Networks (ANN-PSR) are then compared by one-month-ahead forecast 14 using Cuntan and Danjiangkou data. The KNN model performs the best among the four models, but only exhibits 15 weak superiority to ARMA. Further analysis demonstrates that a low correlation between model inputs and 16 outputs could be the main reason to restrict the power of ANN. A Moving Average Artificial Neural Networks 17 (MA-ANN), using the moving average of streamflow series as inputs, is also proposed in this study. The results 18 show that the MA-ANN has a significant improvement on the forecast accuracy compared with the original four 19 models. This is mainly due to the improvement of correlation between inputs and outputs depending on the 20 moving average operation. The optimal memory lengths of the moving average were three and six for Cuntan and 21 Danjiangkou respectively when the optimal model inputs are recognized as the previous twelve months. 22 23