Resumen
This study develops hourly water level forecasting models with lead-times of 1 to 3 h using an artificial neural network (ANN) for Anyangcheon stream, one of the major tributaries of the Han River, South Korea. To consider the backwater effect from this river, an enhanced tributary water level forecasting model is proposed by adding multiple water level data on the main river as input variables into the conventional ANN structure which often uses rainfall and upstream water level data. Four types of ANN models per each lead-time are built with increasing complexity of the input vector, and their performances are compared. The results indicate that the inclusion of multiple water level data on the main river to the network provides water level forecasts with greater accuracy at the Ogeumgyo gauging station of interest. The final best ANN models for water level forecasts with lead-times of 1 to 2 h show good performance with root mean square errors (RMSE) below 0.06 m and 0.12 m, respectively. However, the final best ANN model for forecasting 3 h ahead was unsatisfactory, showing underestimation at many rising parts of the hydrograph.