Prediction of Discharge in a Tidal River Using Artificial Neural Networks

Discharge predictions at tidally affected river reaches are currently still a great challenge in hydrological practices. In tidal rivers, water levels are not uniquely a function of streamflow. Here, the possibility to predict discharge from water level information from gauge stations at sea and in the river is explored. A hindcast model is established for a tide-dominated lowland site along the Mahakam River (East Kalimantan, Indonesia), using an artificial neural network (ANN) model. The input data for the ANN are gradually increased, by adding new input data in each step. The results show that the inclusion of data from tide predictions at sea leads to an improved model performance. The optimized ANN-based hindcast model produces a good discharge estimation, as shown by a consistent performance during both the training and validation periods. Using this model, discharge can be predicted from astronomical tidal predictions at sea plus water level measurements from a single station at an upstream location. Alternatively, the ANN model can be used as a tool for data gap filling in a disrupted discharge time-series based on a horizontal acoustic Doppler current profiler (H-ADCP). A forecast model is developed for the same river site that is located near the city of Samarinda. To this end, water level data, predicted tide levels, and at-site historical data are considered as input for the model. The discharge time-series derived from H-ADCP data are used for calibration and validation of the multistep ahead discharge forecast model. A good performance is obtained for predictions with a forecast lead time up to two days. (C) 2014 American Society of Civil Engineers.