Although development of a turbid river water forecasting method has been driven, that of real-time is
unpracticed until now. The reason is as follows. The correlation between turbidity transport L and river discharge Q
is usually expressed in a rating curve (L=αQ^β, α, β: coefficient). The coefficients can’t be preassigned because of
different values in each flooding condition. Uncertainties in the coefficients often lead to significant differences
between simulation results with the distributed hydrological model and observations, making it difficult to obtain the
reliable model. In this study, we predicted variations of a turbid river water using sequential data assimilation under
the condition of the prefect forecasted rainfall. The method uses sequential data assimilation combined with the
model. The model is used to estimate a turbid river water with a spatial resolution of 1 km grid and a time
resolution of 1 hour. The sequential data assimilation technique aims at accommodating states of the model to
observations. It is motivated to realize an online estimation of model parameter as the coefficients. To attain the
purpose, sequential data assimilation such as Ensemble Kalman Filter (EnKF) and Particle Filter (PF) is applied. Two
algorithms have a similar methodology in view of the ensemble-based method, are used to identify the coefficients.
The case study to the actual field gave following consequences. System and observation noises were set properly
from the identical twin experiment. In conclusion, this study has demonstrated the effectiveness of the method by
EnKF in real-time a turbid river water forecasting.
View full abstract