2nd GLOBAL CONFERENCE on ENGINEERING RESEARCH, Balıkesir, Türkiye, 7 - 10 Eylül 2022, ss.399
Modeling of stream flow has great importance in terms of water resources management
and planning. Hybrid models built with signal processing and machine learning techniques are
becoming popular recently with realistic prediction results. The wavelet transform technique
is a pre-signal processing method that allows the data to be analyzed in more detail by
separating the data into various components. In this study, mother wavelets commonly used in
hydrometeorological studies such as Haar, Daubechies 2, Daubechies 4, Discrete Meyer,
Coiflets 3, Coiflets 5, Symlet 3, Symlet 5 were used with feed-forward backpropagation
neural network (FFBPNN) to determine which mother wavelet is most effective in streamflow
prediction in Amasya. The correlation matrix was used to assess model input combinations.
Inputs that are thought to have an important relationship with the output component are
presented to the model. Precipitation, temperature and past streamflow values were used to
establish the estimation model. During the modeling phase, 70% of the data was divided into
training, 15% validation and 15% testing. Model performances were evaluated according to
mean square error, correlation coefficient and rank analysis. As a result, the best prediction
results were obtained with the Coiflet 5 discrete wavelet. The symlet 3 wavelet showed the
worst results. In addition, all the established Wavelet FFBPNN models except the symlet 3
wavelet were superior to the stand-alone FFBPNN model. The study results are important in
terms of water resources method, sediment control, water structure construction and flood
drought management plan development.