Monthly streamflow prediction in Amasya, Türkiye, using an integrated approach of a feedforward backpropagation neural network and discrete wavelet transform

Due to climate change and increasing demand for water, effective planning of water resources is a current issue. Reliable and accurate streamflow prediction is of great importance in the planning of water resources. This study aimed to predict monthly streamflows in Amasya by combining a discrete wavelet transform and a feedforward backpropagation neural network (FFBPNN) model. Various meteorological variables were separated into sub signals with mother wavelets commonly used in hydrometeorological studies, such as Haar, Daubechies 2, Daubechies 4, Discrete Meyer, Coiflet 3, Coiflet 5, Symlet 3, and Symlet 5, and entered into the FFBBNN model to create a hybrid wavelet-based FFBBNN model. Inputs with a significant relationship with the output were entered into the model. Precipitation, temperature, and previous streamflow values covering 1960–2011 were used to create the model. During the modeling phase, 70% of the data were divided into training, 15% into validation, and 15% into testing. The performance of the model was compared using mean square error, correlation coefficient, and rank analysis. Coiflet 5 mother wavelet showed the best results. Moreover, it was proven that monthly streamflow can be successfully predicted using previous precipitation, temperature, and streamflow values and the Coiflet 5 mother wavelet with the FFBBNN hybrid model (MSE: 7.143, R: 0.921). In addition, all built wavelet FFBPNN models except the Symlet 3 mother wavelet performed better than the single FFBPNN model. The results of the study will assist planners, and decision makers in terms of providing sustainable and effective water resources and drought management. © 2022, The Author(s), under exclusive licence to Springer Nature Switzerland AG.