A hydrologi̇c modelli̇ng based on machi̇ne learni̇ng: firtina stream case

Water is the basis of life on Earth. The lack of water makes it impossible for many of the creatures on Earth to live. So it is very important to be close to water sources since its vital importance. But being close to the water has also brought some problems. The cities set up near water sources hat to cope with some problems such as flood, tidal currents, and changes in the water lines. People whose residential areas and farmland have been destroyed as a result of these problems have been forced to leave their settlements. They had also faced financial and spiritual damages.These losses can be avoided thanks to the accurate flow predictions to be obtained by performing regional analyzes. In this study, flow predictions were carried out through machine learning and deep learning models.Machine learning is a method developed through programming languages that can generate predictions using the learning algorithms. In this study, Xgboost machine learning, LSTM and GRU as deep learning training algorithms were performed and compared. Fırtına River basin has been chosen as the study site. Daily maximum temperature, minimum temperature, relative humidity, solar radiation and wind variables from the CFSR (climate prediction system analysis) system were used for independent variable data sets of the study site. Current observation data were obtained from the Institute of Electrical Affairs (EIS) 2232 current observation station at Topluca Village location. The data covers a total of 8758 days between 1 January 1979 and 31 December 2002. The data were used in prediction models by separating into 21 independent and 1 dependent variable and flow predictions of 875 days were performed. As a result, it was seen that LSTM model had the best performance while XGBoost had the worst. GRU model has shown quite good performance in estimating flow. In addition, it was observed that the CFSR datasets are suitable to use in machine learning hydrological modeling.