Improving numerical forecast accuracy with ensemble Kalman filter and chaos theory: Case study on Ciliwung river model

Abstract

• A new hybrid data assimilation scheme is proposed based on ensemble Kalman filter and chaos theory. • This hybrid data assimilation scheme is applied to improve the forecasting accuracy of the Ciliwung river model. • The numerical model is adapted to simulate a real-time forecast; for which the data assimilation is implemented. • Results from different forecast schemes are produced, and detailed comparisons are conducted. • The hybrid scheme out-performs others with more than 50% of RMSE removed. The classic Kalman filter implementation uses the measurements up to the time of forecast to update the initial conditions of the numerical model, with the updating effect limited to a prediction horizon when the improved initial conditions are washed out. To further enhance the prediction capability, this study proposes a new hybrid data assimilation scheme, which adopts chaos theory to predict the measurements into the forecast phase, and then assimilates the predicted measurements into the numerical model using the ensemble Kalman filter. The hybrid data assimilation scheme is applied in a simulated real-time forecast of the Ciliwung river model. It is revealed that the hybrid scheme can further improve the modelling accuracy up to a prediction horizon of 4 days as compared to the update based solely on the ensemble Kalman filter.