A Comparison of Models for River Flowrate Prediction Considering Water Use

Development of a model for river flowrate prediction in high accuracy is essential for sustainable development of water resources and preservation of riparian ecosystem. Besides, it has become harder and more complex to estimate the water transfer such as the amount of water intake, sewage discharge and precipitation along with the improved quality of life, the technological advancement and a climate change. Thus, the models are based on an artificial neural network (ANN), which is applied in many different fields for predictive modelling. In this study, we compared the accuracy of developed models considering different variables as input data to predict river flowrate. Each model was applied using developed database in a sub-basin of South Korea. The patterns of water use are classified according to the purpose of water use: residual, industrial and agricultural. First model is a typical model based on an ANN using the river flowrate in the past. Second model is based on ANN using the water use including water intake and sewage discharge and the river flowrate in the past. Last one considers water use, flowrate and precipitation in the past as input variables. Especially, water intake was efficiently estimated by using time series decomposition analysis to derive the trend depending on a climate change, a population growth, changes in temperature and economic prosperity and the daily events including water supply cut-off and water supply for drought areas in a short period. Finally, the three models were assessed in the respect of the temporal economic and the accuracy. The accuracy was dependent on residential water and agricultural water. It is suggested that these models are more effectively applicable for regions according to the characteristics of water consumption. This model can be used to determine a criteria for the permission of water intake and distribute river water efficiently in a drought.