INVESTIGATION ON ACCURATE PREDICTION OF WATER LEVEL IN A SMALL-MEDIUM SIZED RIVER BASIN USING DEEP LEARNING

Abstract

 In this study, we developed water level prediction model for small-medium river basin using LSTM based deep learning and examined the accuracy under several different conditions of data type and structure of input and output layers. The analyzed results showed that the Nash-Sutcliff coefficient was more 0.9, and error rate of the peak water level was less 10% by the model with multiple points data set of rainfall and hourly water level change for the input layer and the multiple points water level for the output layer. Moreover, the time difference of peak water level occurrence was smaller than in other calculation cases. Therefore, it is shown that water level prediction model developed in this study can be used to predict the water level 3 hours ahead at multiple stations with high accuracy in the small-medium river basin.