Prediction and comparison of streamflow using hybrid and independent models in Zola dam basin

Abstract

Raise of mean temperature, decrease in rainfall, increase in water consumption, consecutive droughts, climate changes, global warming are among the factors that make water, especially surface water, a valuable resource and its efficient use has become of importance. Therefore, today experts and researchers apply different methods and approaches to model and forecast the changes of this valuable material and are constantly developing and updating these modeling tools. In the present research, extreme learning machine (ELM) and long short-term memory (LSTM) algorithms were implemented to model and forecast the streamflow of the Zola Dam watershed. Also, the methods of wavelet function and variational mode decomposition (VMD) were applied to build hybrid models and increase modeling accuracy. Two error indices, namely root mean square error (RMSE) and mean absolute error (MAE) and Kling–Gupta efficiency performance index were calculated to compare the results and select the best performance. The results showed that the use of hybrid models increase the performance and accuracy of the projections. All in all, the ELM-VMD approach has the best and most accurate performance with the values of evaluation indices RMSE=0.34, MAE=0.36 and KGE=0.992.