Investigation of Impact of Vapor Pressure on Hybrid Streamflow Prediction Modeling

Abstract

In this study, daily streamflow prediction models have been developed for Aksu Stream, in the Eastern Black Sea Basin of Turkey. To reach at this aim, hybrid artificial intelligence models have been developed, by using a new parameter, vapor pressure. Vapor pressure efficiency has been investigated for hybrid streamflow prediction models. Streamflow prediction models have been developed by using Artificial Neural Network (ANN), Multivariate Adaptive Regression Splines (MARS), and their hybrid models. Hybridization of streamflow prediction models has been made with Wavelet Transform (WT). 10 yearly daily hydrological (discharge (m(3)/s)), meteorological (precipitation (mm), vapor pressure (hPA)) data, and seasonality coefficient have been used as input data of streamflow prediction models. In the selection of the best streamflow prediction model, 14 different day-delayed input combinations have been established by using 10 yearly data. As a result of the study, the highest flow forecast performance model has been determined as Wavelet Artificial Neural Network (WANN) in the study area. In the WANN model, the vapor pressure parameter was found to reduce the error by about 18.5% and improve the forecast performance. This study has concluded that, vapor pressure may be used in the future studies as a new parameter for streamflow prediction models.