COMBINED NEURAL NETWORK AND ENSEMBLE KALMAN FILTER APPLICATION FOR DISCHARGE AND WATER LEVEL FORECASTING IN THE RIVER RHINE

摘要

Numerical simulation models often require considerable computation time which may hamper their use in real-time control and forecasting applications. Artificial Neural Networks (ANN) are known to have the ability to emulate a well-tuned numerical model, be it at a limited number stations. In this paper, a combination of neural network and data assimilation techniques is explored. The ANN is used to mimic the behaviour of the numerical model at location(s) of interest and then the Ensemble Kalman filter (EnKF) is applied on the neural network structure. This will decrease the computational burden of applying data assimilation directly on the numerical model itself. The application presented in this paper is for discharge and water level forecasting at Lobith, which is a station on the river Rhine at the border of The Netherlands.