Conjunction Model Design for Intermittent Streamflow Forecasts: Extreme Learning Machine with Discrete Wavelet Transform

Forecasting intermittent streamflow is essential for water management and water quality, planning of water supply, hydropower and irrigation systems. This chapter proposes a novel intelligent data analytic method using extreme learning machines combined with discrete wavelet transform (ELM-DWT) to forecast intermittent streamflow. Daily streamflow data from Uzunkopru and Babaeski, Turkey are utilized. Model input combinations involving antecedent flow are considered. The results are compared with artificial neural network combined with discrete wavelet transform (ANN-DWT) and single ELM and ANN methods. Optimal input combinations are decided utilizing auto-correlation and partial auto-correlations. Models are evaluated based on the three measures, root mean square errors (RMSE), Nash-Sutcliffe efficiency (NSE) and coefficient of correlation (R). In both stations, the optimal ELM-DWT model with five previous flow values as inputs provided the best accuracy compared to other corresponding models. DWT considerably increased the accuracy of single models in forecasting intermittent streamflow of both stations. The RMSE of the best ELM and ANN models were decreased by 38 (61)% and 59 (69)% using the proposed model (ELM-DWT) for the Uzunkopru (Babaeski) stations, respectively. The chapter demonstrates the importance of conjunction model design for intermittent streamflow forecasting with extreme learning machines and discrete wavelet transform as potential intelligent modeling methods for natural hazard and other risk mitigation tools.