Improving the accuracy of rainfall-runoff relationship estimation using signal processing techniques, bio-inspired swarm intelligence and artificial intelligence algorithms

In water resources engineering, streamflow estimation models are of great importance. The use of black box models in determining streamflow estimation is preferred because it saves time compared to deterministic models. In addition, the data needed is less in quality and quantity and has a lower transaction volume. Therefore, accurate flow estimation plays a crucial role in water resources planning, management, and sizing of water structures. This study aimed to forecast the monthly streamflow using particle swarm optimization, least-squares support-vector machine, and signal processing models. For this purpose, 575 monthly data between 1963-2010 were used in Sivas, and 622 monthly data between 1960-2011 were used in Kayseri. The model structure used past streamflow, temperature, and precipitation values as input. During the installation of hybrid models, the inputs are divided into subcomponents with three levels of discrete Meyer mother wavelet and presented to the LS-SVM model. The performance of the models was evaluated with the help of statistical indicators such as determination coefficient, mean absolute error, Nash–Sutcliffe Efficiency, mean square error, Violin plots and visual Taylor diagrams. At the end of the study, EMD-LSSVM was the best streamflow estimation model at station 1501, and W-LSSVM was the most successful streamflow estimation model at station 1535. © 2023, The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature.